E PRELIMINARY
Information in this document is provided in connection with Intel products. No license, express or implied, by estoppel or otherwise, to any intellectual property rights
is granted by this document or by the sale of Intel products. Except as provided in Intel’s Terms and Conditions of Sale for such products, Intel assumes no liability
whatsoever, and Intel disclaims any express or implied warranty, relating to sale and/or use of Intel products including liability or warranties relating to fitness for a
particular purpose, merchantability, or infringement of any patent, copyright or other intellectual property right. Intel products are not intended for use in medical, life
saving, or life sustaining applications. Intel retains the right to make changes to specifications and product descriptions at any time, without notice. The Intel 430VX
PCIset may contain design defects or errors known as errata. Current characterized errata are available on request. *Third-party brands and names are the property
of their respective owners.
© INTEL CORPORATION 1996 July 1996 Order Number: 290553-001
Supports All 3V Pentium Processors
PCI 2.1 Compliant
Integrated DRAM Controller
− − 64-Bit Path to Memory
− − 4 MB to 128 MB of Main memory
− − EDO/Fast Page Mode DRAM Support
(6-2-2-2 Reads at 66 MHz)
− − 5 RAS Lines
− − Support for Symmetrical and
Asymmetrical DRAMs
− − Supports SDRAM (x-1-1-1 at 66 MHz)
− − Buffering for 3-1-1-1 Posted Writes,
DWord and Burst Merging
− − Supports Mixed Memory
Technologies (EDO/SPM/SDRAM)
− − Supports 3V or 5V DRAMs
− − External Buffers on MA Lines are Not
Required
Integrated Second Level Cache
Controller
− − Direct Mapped Organization
− − Supports 256-KB and 512-KB
Pipelined Burst, DRAM Cache and
Standard SRAM
− − Cache Hit Read/Write Cycle Timings
at 3-1-1-1 with Burst or DRAM Cache
SRAM
− − Back-to-Back Read/Write Cycles
at 3-1-1-1-1-1-1-1
− − Supports Write-Back
Fully Synchronous 25/30/33 MHz PCI
Bus Interface
− − 5 PCI Bus Masters (including PIIX3)
− − Converts Back-to-Back sequential
PCI Memory Writes to PCI Burst
Writes
− − CPU-to-PCI Memory Writes Posting
− − PCI-to-DRAM Read Prefetching
− − PCI-to-DRAM Posting
− − Multi-Transaction Timer to Support
Multiple Short PCI Transactions
Shared Memory Buffer Architecture
(SMBA) Support
− − Support Graphics Controller through
a 2-Wire Protocol
− − Supports 1 Row of DRAM
(EDO/FPM/SDRAM)
− − Enhanced Performance Features
Specific to SMBA
Supports the Universal Serial Bus
(USB)
208-Pin QFP System Controller (TVX),
two 100-Pin QFP Data Paths (TDX)
The Intel 430VX PCIset consists of the 82437VX System Controller (TVX), two 82438VX Data Paths (TDX), and
the PCI ISA IDE Xcelerator (PIIX3). The PCIset forms a Host-to-PCI bridge and provides the second level cache
control and a full function 64-bit data path to main memory. The TVX integrates the cache and main memory
DRAM control functions and provides bus control for transfers between the CPU, cache, main memory, and the
PCI Bus. The second level (L2) cache controller supports a write-back cache policy for cache sizes of
256 Kbytes and 512 Kbytes. Cacheless designs are also supported. The cache memory can be implemented
with standard, pipelined burst, or DRAM Cache SRAMs. An external Tag RAM is used for the address tag and
an internal Tag RAM for the cache line status bits. For the TVX's DRAM controller, five rows are supported for
up to 128 Mbytes of main memory. The Shared Memory Buffer Architecture (SMBA) 2-wire interface allows a
graphics controller to use an area of system memory as its frame buffer. The Intel 430VX PCIset has been
enhanced through additional buffers, programmable timers, and burst and DWord merging and optimized DRAM
timings to maintain a high level of performance when used in a SMBA environment. Using the snoop ahead
feature, the TVX allows PCI masters to achieve full PCI bandwidth. The TDXs provide the data paths between
the CPU/cache, main memory, and PCI. For increased system performance, the TDXs contain read prefetch
and posted write buffers.
INTEL 430VX PCISET
82437VX SYSTEM CONTROLLER (TVX) AND
82438VX DATA PATH UNIT (TDX)
82437VX (TVX) AND 82438VX (TDX) E
2
PRELIMINARY
BE[7:0]#
DEVSEL#
PAR
REQ[2:0]#
PHLD#
PHLDA#
GNT[2:0]#
LOCK#
STOP#
IRDY#
TRDY#
FRAME#
C/BE[3:0]#
MSTB[1:0]
MADV#
PCMD[1:0]
HOE#
MOE#
POE#
PLINK[15:0]
Clocks
Host
Interface
D/C#
HLOCK#
HITM#
EADS#
BRDY#
BOFF#
AHOLD
NA#
KEN#/INV
CACHE#
SMIACT#
ADS#
W/R#
M/IO#
TWE#
CCS#/CAB4
COE#
GWE[7:0]#
CADS#/CAA3
CADV#/CAA4
TIO[7:0]
PCLKIN
HCLKIN
RAS[3:0]#/CS[3:0]#
CAS[7:0]#/DQM[7:0]
MA[11:0]
DR AM
Interface
TDX
Interface
PCI
Interface
Cache
Interface
A[31:3]
SRASB#/RAS4#
SCAS[B:A]
AD[31:0]
KRQAK/CAB3
WE[B:A]#
BWE/CGCS#
REQ3#/MREQ#
GNT3#/MGNT#
MXS/CS4#
SRASA#
M
U
X
055301
82437VX TVX Simplified Block Diagram
E82437VX (TVX) AND 82438VX (TDX)
3
PRELIMINARY
MSTB[1:0]
MADV#
PCMD[1:0]
HOE#
MOE#
POE#
PLINK[7:0]
Clocks
Host
Interface
HCLK
MD[31:0]DRAM
Interface
TVX
Interface
D[31:0]
MXS
055302
82438VX TDX Simplified Block Diagram
82437VX (TVX) AND 82438VX (TDX) E
4
PRELIMINARY
CONTENTS
PAGE
1.0. ARCHITECTURE OVERVIEW OF TSC/TDP..................................................................................................7
2.0. SIGNAL DESCRIPTION................................................................................................................................10
2.1. TVX Signals ................................................................................................................................................10
2.1.1. HOST INTERFACE .............................................................................................................................10
2.1.2. DRAM INTERFACE.............................................................................................................................12
2.1.3. SECONDARY CACHE INTERFACE ..................................................................................................12
2.1.4. PCI INTERFACE .................................................................................................................................14
2.1.5. TDX INTERFACE................................................................................................................................15
2.1.6. CLOCK, RESET, AND TEST ..............................................................................................................15
2.2. TDX Signals................................................................................................................................................16
2.2.1 Data Interface Signals...........................................................................................................................16
2.2.1. DATA INTERFACE SIGNALS.............................................................................................................16
2.2.2. TVX INTERFACE SIGNALS................................................................................................................16
2.2.3. CLOCK SIGNAL..................................................................................................................................16
2.3. TVX STRAPPING PINS .............................................................................................................................17
2.4. SIGNAL STATES DURING A HARD RESET............................................................................................18
3.0. REGISTER DESCRIPTION ...........................................................................................................................19
3.1. I/O Mapped Registers.................................................................................................................................19
3.1.1. CONFADDCONFIGURATION ADDRESS REGISTER..................................................................19
3.1.2. CONFDATACONFIGURATION DATA REGISTER........................................................................20
3.2. PCI Configuration Registers .......................................................................................................................21
3.2.1. VIDVENDOR IDENTIFICATION REGISTER..................................................................................22
3.2.2. DIDDEVICE IDENTIFICATION REGISTER....................................................................................23
3.2.3. PCICMDPCI COMMAND REGISTER.............................................................................................23
3.2.4. PCISTSPCI STATUS REGISTER...................................................................................................24
3.2.5. RIDREVISION IDENTIFICATION REGISTER................................................................................24
3.2.6. CLASSCCLASS CODE REGISTER................................................................................................25
3.2.7. HEDTHEADER TYPE REGISTER..................................................................................................25
3.2.8. MLTMASTER LATENCY TIMER REGISTER.................................................................................25
3.2.9. BISTBIST REGISTER......................................................................................................................26
3.2.10. ACONARBITRATION CONTROL REGISTER..............................................................................26
3.2.11. PCONPCI CONTROL....................................................................................................................27
3.2.12. CCCACHE CONTROL REGISTER...............................................................................................27
3.2.13. CCECACHE CONTROL EXTENDED REGISTER.......................................................................29
3.2.14. SDRAMCSDRAM CONTROL REGISTER....................................................................................30
3.2.15. DRAMECDRAM EXTENDED CONTROL REGISTER.................................................................31
3.2.16. DRAMCDRAM CONTROL REGISTER.........................................................................................32
3.2.17. DRAMTDRAM TIMING REGISTER ..............................................................................................33
E82437VX (TVX) AND 82438VX (TDX)
5
PRELIMINARY
3.2.18. PAMPROGRAMMABLE ATTRIBUTE MAP REGISTERS (PAM[6:0])..........................................34
3.2.19. DRBDRAM ROW BOUNDARY REGISTERS...............................................................................37
3.2.20. DRTHDRAM ROW TYPE REGISTER HIGH................................................................................39
3.2.21. DRTLDRAM ROW TYPE REGISTER LOW..................................................................................40
3.2.22. TRDTPCI TRDY TIMER ................................................................................................................41
3.2.23. MTTMULTI-TRANSACTION TIMER REGISTER..........................................................................41
3.2.24. SMRAMSYSTEM MANAGEMENT RAM CONTROL REGISTER................................................42
3.2.25. SMBCRSMB CONTROL REGISTER............................................................................................43
3.2.26. SMBSASMB START ADDRESS...................................................................................................44
3.2.27. GCLTGRAPHICS CONTROLLER LATENCY TIMER ..................................................................44
4.0. FUNCTIONAL DESCRIPTION......................................................................................................................45
4.1. Host Interface..............................................................................................................................................45
4.2. PCI Interface...............................................................................................................................................45
4.3. Secondary Cache Interface........................................................................................................................45
4.3.1. CLOCK LATENCIES ...........................................................................................................................49
4.3.2. SNOOP CYCLES ................................................................................................................................49
4.4. DRAM Interface ..........................................................................................................................................50
4.4.1. DRAM ORGANIZATION......................................................................................................................51
4.4.2. DRAM ADDRESS TRANSLATION.....................................................................................................57
4.4.3. DRAM PAGING...................................................................................................................................58
4.4.4. EDO MODE .........................................................................................................................................58
4.4.5. SDRAM MODE....................................................................................................................................58
4.4.6. AUTO DETECTION.............................................................................................................................59
4.4.7. DRAM PERFORMANCE.....................................................................................................................59
4.4.8. DRAM REFRESH................................................................................................................................62
4.4.9. SYSTEM MANAGEMENT RAM..........................................................................................................62
4.5. TDXs and the PLINK Interface ...................................................................................................................62
4.6. System Arbitration.......................................................................................................................................63
4.6.1. PRIORITY SCHEME AND BUS GRANT............................................................................................64
4.6.2. MULTI-TRANSACTION TIMER (MTT)................................................................................................65
4.6.3. CPU POLICIES....................................................................................................................................65
4.7. Shared Memory Buffer Architecture ...........................................................................................................65
4.7.1. SMB SYSTEM MEMORY MAPPING..................................................................................................65
4.7.2. CPU ACCESS TO SMB ......................................................................................................................67
4.7.3. PCI MASTER ACCESS TO SMB........................................................................................................67
4.7.4. GC (GRAPHICS CONTROLLER) ACCESS TO SMB........................................................................67
4.7.5. SMB INTERFACE................................................................................................................................68
4.7.6. SMBA ARBITRATION .........................................................................................................................70
4.7.7. TVX-TO-GC LATENCY CONTROL MANAGEMENT.........................................................................70
4.8. Clock Generation and Distribution..............................................................................................................71
4.8.1. RESET SEQUENCING .......................................................................................................................71
82437VX (TVX) AND 82438VX (TDX) E
6
PRELIMINARY
5.0. PINOUT AND PACKAGE INFORMATION...................................................................................................72
5.1. TVX Pinout..................................................................................................................................................72
5.2. TDX Pinout..................................................................................................................................................76
5.3. TVX Package Dimensions..........................................................................................................................79
5.4. TDX Package Dimensions..........................................................................................................................80
6.0. TESTABILITY.................................................................................................................................................81
6.1. 82437VX TVX Testability............................................................................................................................81
6.1.1. TVX NAND TREE MODE....................................................................................................................81
6.1.2. TVX ID CODE EXTRACTION.............................................................................................................87
6.1.3. TVX TRI-STATE MODE ......................................................................................................................87
6.2. 82438VX TDX Testability............................................................................................................................87
6.2.1. TDX NAND TREE MODE....................................................................................................................88
6.2.2. TDX ID CODE EXTRACTION.............................................................................................................90
6.2.3. TDX TRI-STATE MODE......................................................................................................................90
E82437VX (TVX) AND 82438VX (TDX)
7
PRELIMINARY
1.0. ARCHITECTURE OVERVIEW OF TSC/TDP
The Intel 430VX PCIset (Figure 1) consists of the 82437VX System Controller (TVX), two 82438VX Data Path
(TDX) units, and the PCI IDE ISA Xcelerator (PIIX3). The TVX and two TDXs form a Host-to-PCI bridge. The
PIIX3 is a multi-function PCI device providing a PCI-to-ISA bridge, a fast IDE interface, an APIC interface, and a
host/hub controller for the Universal Serial Bus (USB). The PIIX3 also provides power management.
The two TDXs provide a 64-bit data path to the host and to main memory and provide a 16-bit data path (PLINK)
between the TVX and TDX. PLINK provides the data path for CPU to PCI accesses and for PCI to main memory
accesses. The TVX and TDX bus interfaces are designed for 3V and 5V busses. The Intel 430VX PCIset
connects directly to the Pentium® processor 3V host bus; The Intel 430VX PCIset connects directly to 5V or 3V
main memory DRAMs; and the TVX connects directly to the 5V PCI Bus.
The TVX and TDX interface with the Pentium processor host bus, a dedicated memory data bus, and the PCI
bus. The Intel 430VX PCIset implements a Shared Memory Buffer Architecture (SMBA) handshake 2-wire
protocol that allows a graphics controller to use a portion of system memory as its frame buffer region. In
addition, the PLINK bus is used to connect the PCI bus with the TDX, through the TVX (see Figure 1).
DRAM Interface
The DRAM interface is a 64-bit data path that supports Standard Page Mode (SPM), Extended Data Out (EDO),
and Synchronous DRAM (SDRAM) memory. The DRAM interface supports 4 Mbytes to 128 Mbytes of system
memory with five RAS lines and also supports symmetrical and asymmetrical addressing for 512Kx32, 1Mx32,
2Mx32, and 4Mx32 deep SIMM modules (single- and double- sided). The TVX supports SDRAM 1Mx 64, 2Mx
64, and 4Mx 64 deep DIMM modules (asymmetrical single- and double- sided). The Intel 430VX PCIset does not
support parity and requires that x32 and x64 SIMMs/DIMMs be used.
Second Level Cache
The TVX supports a write-back cache policy providing all necessary snoop functions and inquire cycles. The
second level cache is direct mapped and supports both a 256-Kbyte or 512-Kbyte SRAM configuration using
pipelined burst, DRAM Cache, or standard SRAMs. DRAM Cache is a DRAM based cache alternative to
piplined burst SRAM. Its pinout is a superset of pipeline burst and conforms to the standard pipeline burst
footprint. One chipset signal (KRQAK), two system signals (H/WR# and RESET#), and one DRAM Cache
specific signal (M/S#) are the only signal differenences between pipeline burst SRAM and DRAM Cache. The
Pipeline burst or DRAM Cache configuration performance is 3-1-1-1 for read/write cycles; back-to-back reads
can maintain a 3-1-1-1-1-1-1-1 transfer rate.
TDX
Two TDXs create a 64-bit CPU memory data path. The TDXs also interface to the 16-bit PLINK inter-chip bus on
the TVX for PCI transactions. The combination of the 64-bit memory path and the 16-bit PLINK bus make the
TDXs a cost effective solution, providing optimal CPU-to-main memory performance, while maintaining a small
package footprint (100 pins each).
82437VX (TVX) AND 82438VX (TDX) E
8
PRELIMINARY
ISA Bus
Pentium Processor
TVX
Main
Memory
(DRAM)
Address
Data
Control
TDX
Data
Addr
Cntl
TXD Cntl
Control
Address/Data
PCI Bus
Host Bus
Cntl
PCI
Device(s)
Cache
(SRAM)
®
Second Level
Cache
Tag Cntl
TIO[7:0] PLINK (Data)
ISA
Device(s)
Fast IDE
Hard
Disk
CD ROM
PIIX3
Tag
Graphics
2
Arb
USB
1USB
2
Universal Serial Bus
3.3V
055303
Figure 1. Intel 430VX PCIset System Block Diagram
E82437VX (TVX) AND 82438VX (TDX)
9
PRELIMINARY
PCI Interface
The PCI interface is 2.1 compliant and supports up to four PCI bus masters in addition to the PIIX3 bus master
requests. The TVX and TDXs together provide the interface between PCI and main memory; however, only the
TVX connects to the PCI bus.
Buffers
The TVX and TDXs together contain three sets of buffers for optimizing data flow. A DRAM write buffer is
provided for CPU-to-main memory writes, second level cache write-back cycles, and PCI-to-main memory
transfers. This buffer is used to achieve 3-1-1-1 posted writes to main memory, and also provides DWord
merging and burst merging for CPU to main memory write cycles. Buffering is provided for PCI-to-main memory
writes. A buffer is provided for CPU-to-PCI writes to maximize the bandwidth for graphic writes to the PCI bus in
a non-SMBA system. In addition, PCI-to-main memory read pre-fetch buffering permits up to two lines of data to
be prefetched at an x-2-2-2 rate.
Shared Memory Buffer Architecture (SMBA)
The Intel 430VX PCIset provides a hardware interface that allows a graphics controller to access an area of
system memory as a frame buffer. This reduces system cost by eliminating the need to have separate memory
for the graphics subsystem. Two signals are used to arbitrate ownership of DRAM (DRAM address and control
signals). The Intel 430VX PCIset has been enhanced as follows to maintain a high level of performance when
used in a SMBA environment:
•Buffering for improved CPU and PCI posting and PCI pre-fetching
•Programmable timers to maximize performance and establish a balance between the graphics controller and
the system (regulates read and write accesses to DRAM)
•Burst merging and DWord merging for efficient DRAM writes
•Optimized DRAM Read timings
System Clocking
The processor, secondary cache, main memory subsystem, and PLINK bus all run synchronously to the host
clock. The host clock frequencies supported are 50 MHz, 60 MHz, and 66 MHz. The PCI clock runs
synchronously at half the host clock frequency. The TVX and TDXs have a host clock input and the TVX has a
PCI clock input. These clocks come from an external source and have a maximum clock skew requirement with
respect to each other.
82437VX (TVX) AND 82438VX (TDX) E
10
PRELIMINARY
2.0. SIGNAL DESCRIPTION
This section provides a detailed description of each signal. The signals are arranged in functional groups
according to their associated interface.
The “#” symbol at the end of a signal name indicates that the active, or asserted state occurs when the signal is
at a low voltage level. When ”#” is not present after the signal name, the signal is asserted when at the high
voltage level.
The terms assertion and negation are used extensively. This is done to avoid confusion when working with a
mixture of ”active-low” and ”active-high” signals. The term assert, or assertion indicates that a signal is active,
independent of whether that level is represented by a high or low voltage. The term negate, or negation
indicates that a signal is inactive.
In the Type column 3V/5V indicates that the signal drives 3.3V levels as an output and is 5V tolerant as an input.
All 3V outputs can drive 5V TTL inputs. PCI signals are compliant with the PCI 5.0V Signaling Environment DC
and AC Specifications.
The following notations are used to describe the signal and type:
IInput pin
OOutput pin
OD Open Drain Output pin. This requires a pull-up to the VCC of the processor core
I/O Bi-directional Input/Output pin
2.1. TVX Signals
2.1.1. HOST INTERFACE
Name
Type
Description
A[31:3]
I/O
3V
Address Bus. A[31:3] connects to the address bus of the CPU. During CPU
cycles A[31:3] are inputs. The TVX drives A[31:3] during inquire cycles on behalf of
PCI initiators. A27 has an internal pulldown resistor.
BE[7:0]#
I
3V
Byte Enables. The CPU byte enables indicate which byte lane the current CPU
cycle is accessing. All eight byte lanes must be provided to the CPU if the cycle is
a cacheable read, regardless of the state of BE[7:0]#.
ADS#
I
3V
Address Status. The CPU asserts ADS# to indicate that a new bus cycle is being
driven.
BRDY#
O
3V
Bus Ready. The TVX asserts BRDY# to indicate to the CPU that data is available
on reads or has been received on writes.
NA#
O
3V
Next Address. When burst SRAMs are used in the second level cache or the
second level cache is disabled, the TVX asserts NA# in T2 during CPU write
cycles and with the first assertion of BRDY# during CPU line fills. NA# is never
asserted if the L2 cache is enabled with asynchronous SRAMs.
AHOLD
O
3V
Address Hold. The TVX asserts AHOLD when a PCI initiator is performing a cycle
to DRAM. AHOLD is held for the duration of the PCI burst transfer. The TVX
negates AHOLD when the completion of the PCI to DRAM read or write cycles
complete and during PCI peer transfers.
EADS#
O
3V
External Address Strobe. Asserted by the TVX to inquire the first level cache
when servicing PCI master references to main memory.
E82437VX (TVX) AND 82438VX (TDX)
11
PRELIMINARY
Name
Type
Description
BOFF#
O
3V
Back Off. Asserted by the TVX when required to terminate a CPU cycle that was in
progress.
HITM#
I
3V
Hit Modified. Asserted by the CPU to indicate that the address presented with the
last assertion of EADS# is modified in the first level cache and needs to be written
back.
M/IO#,
D/C#, W/R#
I
3V
Memory/IO; Data/Control; Write/Read. Asserted by the CPU with ADS# to
indicate the type of cycle that the system needs to perform.
HLOCK#
I
3V
Host Lock. All CPU cycles sampled with the assertion of HLOCK# and ADS#, until
the negation of HLOCK# must be atomic (i.e., no PCI activity to main memory is
allowed).
CACHE#
I
3V
Cacheable. Asserted by the CPU during a read cycle to indicate the CPU will
perform a burst line fill. Asserted by the CPU during a write cycle to indicate the
CPU will perform a burst write-back cycle. If CACHE# is asserted to indicate
cacheability, the TVX will assert KEN# either with the first BRDY#, or with NA# if
NA# is asserted before the first BRDY#.
KEN#/INV
O
3V
Cache Enable. KEN#/INV functions as both the KEN# signal during CPU read
cycles, and the INV signal during L1 snoop cycles. During CPU cycles, KEN#/INV
is normally low. KEN#/INV will be driven high during the 1st BRDY# or NA#
assertion of a
non-L1-cacheable
CPU read cycle.
KEN#/INV is driven high(low) during the EADS# assertion of a PCI master DRAM
write(read) snoop cycle. Note that KEN#/INV operation during snoop cycles is
independent of the FLCE bit programming.
SMIACT#
I
3V
System Management Interrupt Active. SMIACT# is asserted by the CPU when it
is in system management mode as a result of an SMI. This signal must be sampled
active with ADS# for the processor to access the SMM space of DRAM, located at
A0000h, after SMM space has been loaded and locked by BIOS at system boot.
NOTES:
All of the signals in the host interface are described in the Pentium processor data sheet. The preceding table highlights Intel
430VX PCIset specific uses of these signals.
82437VX (TVX) AND 82438VX (TDX) E
12
PRELIMINARY
2.1.2. DRAM INTERFACE
Name
Type
Description
RAS[3:0]#/
CS[3:0]#
O
3V
Row Address Strobe(EDO/SPM). RAS[3:0]# select the DRAM row.
Chip Select (SDRAM). CS[3:0]# activate the SDRAM to accept any command
when it is low.
MXS/CS4#
O
3V
Chip Select 4 (SDRAM). CS4# is multiplexed with the MXS TVX-to-TDX interface
signal. If SDRAM is detected in the 5th ROW at power up, MXS/CS4# becomes
CS4#. If EDO/SPM DRAM or no DRAM is detected in the 5th Row, MXS/CS4#
becomes MXS. Note: if 5 rows of SDRAM is implemented, MXS at the TDX must
be tied high with a 1 K
Ω
pullup resistor.
CAS[7:0]#/
DQM[7:0]
O
3V
Column Address Strobe(EDO/SPM). CAS[7:0]# select which bytes are affected
by a DRAM cycle. These signals have internal pullup resistors.
Input/Output Data Mask (SDRAM). DQM[7:0] act as synchronized output enables
during a read cycle and a byte mask during a write cycle. The read cycles require
Tdqz clock latency before the functions are actually performed. In case of a write
cycle, word mask functions are performed in the same cycle (0 clock latency).
SRASA#
SRASB#/
RAS4#
O
3V
SDRAM Row Address Strobe (SDRAM)/Row address 4. SRAS[B:A]# latch row
address on the positive edge of the clock with SRAS[B:A]# low. These signals
enable row access and Precharge. Two copies are provided for loading purposes.
If EDO or SPM DRAM is detected in the 5th ROW at power up, SRASB#/RAS4#
becomes RAS4#. If SDRAM or no DRAM is detected in the 5th Row,
SRASB#/RAS4# becomes SRASB#.
SCAS[B:A]
#
O
3V
SDRAM Column Address Strobe (SDRAM). SCAS[B:A]# latch column address
on the positive edge of the clock with SCAS[B:A]# low. These signals enable
column access. Two copies provided for loading purposes.
MA[11:0]
O
3V
Memory Address (EDO/SPM/SDRAM). This is the row and column address for
DRAM.
WE[B:A]#
O
3V
Write Enable(EDO/SPM/SDRAM). This is the write enable pin for the DRAMs.
Two copies provided for loading purposes. This signal has an internal pullup
resistor.
2.1.3. SECONDARY CACHE INTERFACE
Name
Type
Description
CADV#/
CAA4
O
3V
Cache Advance/Cache Address 4 (copy A). This pin has two modes. The
CADV# mode is used when the second level cache consists of burst SRAM's. In
this mode assertion causes the BSRAM in the secondary cache to advance to the
next QWord in the cache line. The CAA4 mode is used when the second level
cache consists of standard asynchronous SRAMs. In this mode the signal is used
to sequence through the QWords in a cache line during a burst operation.
E82437VX (TVX) AND 82438VX (TDX)
13
PRELIMINARY
Name
Type
Description
CADS#/
CAA3
O
3V
Cache Address Strobe/Cache Address 3 (copy A). This pin has two modes. The
CADS# mode is used when the second level cache consists of burst SRAMs. In
this mode assertion causes the BSRAM in the secondary cache to load the
BSRAM address register from the BSRAM address pins. The CAA3 mode is used
when the second level cache consists of standard asynchronous SRAMs. In this
mode the signal is used to sequence through the QWords in a cache line during a
burst operation.
KRQAK/
CAB3
I/O
3V
DRAM Cache Refresh Request and Acknowledge/Cache Address 3 (copy B).
This pin has two modes. One mode is used when the DRAM Cache SRAM is in the
system. In this mode, KRQAK is a bidirectional refresh request/acknowledge. The
CAB3 function is used when the second level cache consists of standard
asynchronous SRAMs. In this mode, CAB3 is used to sequence through the
QWords in a cache line during a burst operation. This signal has an internal
pulldown resistor.
CCS#/
CAB4
O
3V
Cache Chip Select/Cache Address (copy B). A second level cache consisting of
burst SRAMs will power up, if necessary, and perform an access if this signal is
asserted when CADS# is asserted. A second level cache consisting of burst
SRAMs will power down if this signal is negated when CADS# is asserted. When
CCS# is negated a second level cache consisting of burst SRAMs ignores ADS#. If
CCS# is asserted when ADS# is asserted a second level cache consisting of burst
SRAMs will power up, if necessary, and perform an access. CAB4 is used to
sequence through the QWords in a cache line during a burst operation when using
standard asynchronous SRAM.
COE#
O
3V
Cache Output Enable. The secondary cache data RAMs drive the CPUs data bus
when COE# is asserted.
GWE#
O
3V
Global-Write Enable. When the L2 RAM type is Pipelined Burst, GWE# asserted
causes a QWord to be written into the secondary cache data RAMs if they are
powered up. It is used for L2 line fills.
When the L2 RAM type is standard asynchronous SRAM, GWE# is asserted for all
L2 writes, including L2-cacheable CPU writes and L2 line fills.
BWE#/
CGCS#
O
3V
Byte-Write Enable/Cache Global Chip Select. When the L2 RAM type is
Pipelined Burst, this pin functions as a byte write enable. When GWE#=1, the
assertion of BWE# causes the byte lanes that are enabled via the CPU’s
HBE[7:0]# signals to be written to the secondary cache data RAMs, if they are
powered up. Note that HBE[7:0]# are directly connected to the PBSRAMs.
When the L2 RAM type is standard asynchronous RAMs, this pin acts as a global
chip select. CGCS# is asserted for all L2 reads and L2 line fills. For L2 writes, it is
deasserted. Note that some external logic is required for a standard asynchronous
SRAM L2 implementation. See the Secondary Cache Interface section.
TIO[7:0]
I/O
3\5V
Tag Address. These are inputs during CPU accesses and outputs during second
level cache line fills and the second level cache line invalidates due to inquire
cycles. TIO[7:0] contain the second level tag address for a 256-Kbyte second level
caches. TIO[6:0] contains the second level tag address and TIO7 contains the
second level cache valid bit for 512-Kbyte caches. These signals have internal
pulldown resistors.
TWE#
O
3V
Tag Write Enable. When asserted, new state and tag addresses are written into
the external tag.
82437VX (TVX) AND 82438VX (TDX) E
14
PRELIMINARY
2.1.4. PCI INTERFACE
Name
Type
Description
AD[31:0]
I/O
3/5V
Address/Data. The standard PCI address and data lines. Address is driven with
FRAME# assertion, data is driven or received in following clocks.
C/BE[3:0]#
I/O
3/5V
Command/Byte Enable. The command is driven with FRAME# assertion, byte
enables corresponding to supplied or requested data is driven on following clocks.
FRAME#
I/O
3/5V
Frame. Assertion indicates the address phase of a PCI transfer. Negation
indicates that one more data transfer is desired by the cycle initiator.
DEVSEL#
I/O
3/5V
Device Select. This signal is driven by the TVX when a PCI initiator is attempting
to access DRAM. DEVSEL# is asserted at medium decode time.
IRDY#
I/O
3/5V
Initiator Ready. Asserted when the initiator is ready for a data transfer.
TRDY#
I/O
3/5V
Target Ready. Asserted when the target is ready for a data transfer.
STOP#
I/O
3/5V
Stop. Asserted by the target to request the master to stop the current transaction.
LOCK#
I/O
3/5V
Lock. Used to establish, maintain, and release resource locks on PCI.
REQ[2:0]#
I
5V
Request. PCI master requests for PCI.
REQ3#/
MREQ#
I
5V
Request. PCI master requests for PCI. This signal is also multiplexed with the
graphics controller DRAM request line.
GNT[2:0]#
O
3V
PCI Grant. Permission is given to the master to use PCI.
GNT3#/
MGNT#
O
3V
PCI Grant. Permission is given to the master to use PCI. This signal is also
multiplexed with the graphics controller DRAM grant line.
PHLD#
I
5V
PCI Hold. This signal comes from PIIX3. It is the PIIX3 request for PCI.
PHLDA#
O
3V
PCI Hold Acknowledge. This signal is driven by the TVX to grant PCI to PIIX3.
PAR
I/O
3/5V
Parity. A single parity bit is provided over AD[31:0] and C/BE[3:0].
NOTES:
All signals in the PCI interface conform to the PCI Rev 2.1 specification.
E82437VX (TVX) AND 82438VX (TDX)
15
PRELIMINARY
2.1.5. TDX INTERFACE
Name
Type
Description
PLINK[15:0]
I/O
3V
PCI Link. This is the data path between the CPU/DRAM and PCI. Each TDX
connects to one byte of this bus.
MSTB[1:0]
O
3V
Memory Strobe. Controls the data flow and data merging in the DRAM write
buffer.
MADV#
O
3V
Memory Advance. For memory write cycles, assertion causes a QWord to be
drained from the DRAM write buffer and the next data to be made available to the
MD pins of the TDXs. For memory read cycles, assertion causes a QWord to be
latched into the DRAM input register. It is also used to indicate a PCI to DRAM
read.
PCMD[1:0]
O
3V
PLINK Command. This field controls how data is loaded into the DRAM to PCI
buffer, and PLINK input and output registers.
MXS/CS4#
O
3V
Mux Select/Chip Select 4 (SDRAM). When the MXS function is enabled, this
signal provides speed path parameters to the TDX for CPU and PCI accesses
to/from DRAM.
If SDRAM is detected in the 5th ROW at power up, MXS/CS4# becomes CS4#. If
EDO/SPM DRAM or no DRAM is detected in the 5th Row, MXS/CS4# becomes
MXS. Note: if 5 rows of SDRAM is implemented, MXS at the TDX must be tied
high with a 1 K
Ω
pullup resistor.
HOE#
O
3V
Host Output Enable. This signal is used as the output enable for the host data
buus.
MOE#
O
3V
Memory Output Enable. This signal is used as the output enable for the memory
data bus.
POE#
O
3V
PLINK Output Enable. This signal is used as the output enable for the PLINK
Data Bus and selects the DRAM mux and PLINK input mux.
2.1.6. CLOCK, RESET, AND TEST
Name
Type
Description
HCLKIN
I
3V
Host Clock In. This signal pin receives a buffered host clock. This should be the
same clock net that is delivered to the CPU (50 MHz, 60 MHz, and 66 MHz
frequencies are supported). The TVX internal logic uses this signal.
PCLKIN
I
5V
PCI Clock In. This signal pin receives a buffered divide-by-2 host clock. The TVX
internal logic uses this signal.
CPURST
I
3V
Reset. When asserted, this signal asynchronously resets the TVX. The PCI
signals also tri-state compliant to PCI Rev 2.1 specification.
82437VX (TVX) AND 82438VX (TDX) E
16
PRELIMINARY
2.2. TDX Signals
2.2.1 Data Interface Signals
2.2.1. DATA INTERFACE SIGNALS
Name
Type
Description
HD[31:0]
I/O
3V
Host Data. These signals are connected to the CPU data bus. The CPU data bus
is interleaved between the TDX every byte, effectively creating an even and an
odd TDX. The TDXs do not need to know which they are.
MD[31:0]
I/O
3\5V
Memory Data. These signals are connected to the DRAM data bus. The DRAM
data bus is interleaved between the TDXs every byte, effectively creating an even
and an odd TDX. The TDXs do not need to know which they are.
PLINK[7:0]
I/O
3V
PCI Link. These signals are connected to the PLINK data bus on the TVX. This is
the data path between the CPU/DRAM and PCI. PCI to DRAM Reads and CPU to
PCI writes are driven onto these pins by the TDX. CPU to PCI reads and PCI to
DRAM writes are received on this bus by the TDX. Each TDX connects to one
byte of this bus.
2.2.2. TVX INTERFACE SIGNALS
Name
Type
Description
MSTB[1:0]
I
3V
Memory Strobe. See TDX Signals in the TVX Signal description.
MADV#
I
3V
Memory Advance. See TDX Signals in the TVX Signal description.
MXS
I
3V
Mux Select. See TDX Signals in the TVX Signal description.
PCMD[1:0]
I
3V
PLINK Command. See TDX Signals in the TVX Signal description.
HOE#
I
3V
Host Output Enable. See TDX Signals in the TVX Signal description.
MOE#
I
3V
Memory Output Enable. See TDX Signals in the TVX Signal description.
POE#
I
3V
PLINK Output Enable. See TDX Signals in the TVX Signal description.
2.2.3. CLOCK SIGNAL
Name
Type
Description
HCLK
I
3V
Host Clock. Primary clock input used to drive the part.
E82437VX (TVX) AND 82438VX (TDX)
17
PRELIMINARY
2.3. TVX STRAPPING PINS
Name
Pin Name
Description
SCS
A[31:30]
Secondary Cache Size as described in the Cache Control Register (bits [7:6]).
L2RAMT
A[29:28]
Initial L2 RAM Type as described in the Cache Control Register (bits [5:4]).
There are no pullup/pulldown resistors implemented in the A[29:28] I/O buffers.
FD A27
Frequency Detection. BIOS can use this bit to determine if the system is 60
MHz (external pullup) or 66 MHz (no strapping is present) as described in the
DRAM Control Register (bit 0). Bit 0 is initialized with the inverted value of pin A27
after reset negation. The A27 input buffer includes a weak pulldown resistor that
forces DDR[0] to default to 1 if no strapping is present.
DCD KRQAK
DRAM Cache Detection. After reset negation, this bit is sampled to detect if
DRAM cache L2 is present in the system. A 4.7K pull-up must be connected to
the pin, if DRAM cache is in the system. If sampled high, DRAM cache is in the
system. A weak pulldown is provided internally on this pin. A configuration bit (bit
5 in the CCE register, 53h) is provided for BIOS detection.
82437VX (TVX) AND 82438VX (TDX) E
18
PRELIMINARY
2.4. SIGNAL STATES DURING A HARD RESET
The following table shows the state of all TVX and TDX output and bi-directional signals during hard reset
(CPURST asserted).
Table
1
. Signal State During Hard Reset
Signal
State
Signal
State
TVX
TIO[7:0]
Tri-state
A[31:27]
Input
TWE# Low
A[26:3]
Low
AD[31:0]
Low
BRDY#
High
C/BE[3:0]#
Low
NA#
High
FRAME# Tri-state
AHOLD
High
DEVSEL# Tri-state
EADS#
High
IRDY# Tri-state
BOFF#
High
TRDY# Tri-state
KEN#/INV Low STOP# Tri-state
RAS[2:0]#/CS[2:0]#
Low LOCK# Tri-state
RAS3#/CS3# Tri-state
GNT[2:0]#
Tri-state
CAS[7:0]#/DQM[7:0]
Tri-state GNT3#/MGNT# Tri-state
SRASA#
High
PHLDA#
High
SRASB#/RAS4# Tri-state PAR Low
SCASA#
High
PLINK[15:0]
Tri-state
SCASB# Tri-state
MSTB[1:0]
Low
MA[11:0]
Tri-state MADV#
High
WEA#
High
PCMD[1:0]
Low
WEB# Tri-state MXS/CS4# Tri-state
CADV#/CAA4
High
HOE#
High
CADS#/CAA3
High
MOE#
High
CCS#/CAB4 Low POE# Low
KRQAK/CAB3
Input
TDX
COE#
High
HD[31:0]
Tri-state
GWE#
High
MD[31:0]
Tri-state
BWE#/CGCS#
High
PLINK[7:0]
Tri-state
E82437VX (TVX) AND 82438VX (TDX)
19
PRELIMINARY
3.0. REGISTER DESCRIPTION
The TVX contains two sets of software accessible registers (I/O Mapped and Configuration registers), accessed
via the Host CPU I/O address space. The I/O Mapped registers control access to PCI configuration space.
Configuration Registers reside in PCI configuration space and specify PCI configuration, DRAM configuration,
cache configuration, operating parameters, and optional system features.
The TVX internal registers (both I/O Mapped and Configuration registers) are only accessible by the Host CPU
and cannot be accessed by PCI masters. The registers can be accessed as Byte, Word (16-bit), or DWord (32-
bit) quantities, with the exception of CONFADD which can only be accessed as a DWord. All multi-byte numeric
fields use "little-endian" ordering (i.e., lower addresses contain the least significant parts of the field). The
following nomenclature is used for access attributes.
RO Read Only. If a register is read only, writes to this register have no effect.
R/W Read/Write. A register with this attribute can be read and written.
R/WC Read/Write Clear. A register bit with this attribute can be read and written. However, a write of 1
clears (sets to 0) the corresponding bit and a write of 0 has no effect.
Some of the TVX registers described in this section contain reserved bits. Software must deal correctly with
fields that are reserved. On reads, software must use appropriate masks to extract the defined bits and not rely
on reserved bits being any particular value. On writes, software must ensure that the values of reserved bit
positions are preserved. That is, the values of reserved bit positions must first be read, merged with the new
values for other bit positions and then written back.
In addition to reserved bits within a register, the TVX contains address locations in the PCI configuration space
that are marked "Reserved" (Table 1). The TVX responds to accesses to these address locations by completing
the Host cycle. Software should not write to reserved TVX configuration locations in the device-specific region
(above address offset 3Fh).
During a hard reset (CPURST asserted), the TVX sets its internal configuration registers to predetermined
default states. The default state represents the minimum functionality feature set required to successfully bring
up the system. Hence, it does not represent the optimal system configuration. It is the responsibility of the
system initialization software (usually BIOS) to properly determine the DRAM configurations, cache
configuration, operating parameters and optional system features that are applicable, and to program the TVX
registers accordingly.
3.1. I/O Mapped Registers
The TVX contains two registers that reside in the CPU I/O address space—the Configuration Address
(CONFADD) Register and the Configuration Data (CONFDATA) Register. The Configuration Address Register
enables/disables the configuration space and determines what portion of configuration space is visible through
the Configuration Data window.
3.1.1. CONFADDCONFIGURATION ADDRESS REGISTER
I/O Address: 0CF8h (Accessed as a DWord)
Default Value: 00000000h
Access: Read/Write
CONFADD is a 32-bit register accessed only when referenced as a DWord. A Byte or Word reference will "pass
through" the Configuration Address Register to the PCI Bus. The CONFADD Register contains the Bus Number,
Device Number, Function Number, and Register Number for which a subsequent configuration access is
intended.
82437VX (TVX) AND 82438VX (TDX) E
20
PRELIMINARY
Bit
Description
31
Configuration Enable (CONE). 1=Enable. 0=Disable.
30:24
Reserved.
23:16
Bus Number (BUSNUM). When BUSNUM is programmed to 00h, the target of the configuration
cycle is either the TVX or the PCI Bus that is directly connected to the TVX, depending on the
Device Number field. If the Bus Number is programmed to 00h and the TVX is not the target, a
type 0 configuration cycle is generated on PCI. If the Bus Number is non-zero, a type 1
configuration cycle is generated on PCI with the Bus Number mapped to AD[23:16] during the
address phase.
15:11
Device Number (DEVNUM). This field selects one agent on the PCI Bus selected by the Bus
Number. During a Type 1 Configuration cycle, this field is mapped to AD[15:11]. During a Type 0
configuration cycle, this field is decoded and one of AD[31:11] is driven to 1. The TVX is always
Device Number 0.
10:8
Function Number (FUNCNUM). This field is mapped to AD[10:8] during PCI configuration
cycles. This allows the configuration registers of a particular function in a multi-function device to
be accessed. The TVX responds to configuration cycles with a function number of 000b; all other
function number values attempting access to the TVX (Device Number = 0, Bus Number = 0)
generate a type 0 configuration cycle on the PCI Bus with no IDSEL asserted, which results in a
master abort.
7:2
Register Number (REGNUM). This field selects one register within a particular bus, device, and
function as specified by the other fields in the Configuration Address Register. This field is
mapped to AD[7:2] during PCI configuration cycles.
1:0
Reserved.
3.1.2. CONFDATACONFIGURATION DATA REGISTER
I/O Address: 0CFCh
Default Value: 00000000h
Access: Read/Write
CONFDATA is a 32-bit read/write window into configuration space. The portion of configuration space that is
referenced by CONFDATA is determined by the contents of CONFADD.
Bit
Description
31:0
Configuration Data Window (CDW). If bit 31 of CONFADD is 1, any I/O reference in the
CONFDATA I/O space is mapped to configuration space using the contents of CONFADD.
E82437VX (TVX) AND 82438VX (TDX)
21
PRELIMINARY
3.2. PCI Configuration Registers
The PCI Bus defines a slot based "configuration space" that allows each device to contain up to 256 8-bit
configuration registers. The PCI specification defines two bus cycles to access the PCI configuration space
Configuration Read and Configuration Write. While memory and I/O spaces are supported by the Pentium
processor, configuration space is not supported. The PCI specification defines two mechanisms to access
configuration space, Mechanism #1 and Mechanism #2. The TVX only supports Mechanism #1 (both type 0 and
1 accesses). Table 1 shows the TVX configuration space.
The configuration access mechanism makes use of the CONFADD Register and CONFDATA Register. To
reference a configuration register, a DWord I/O write cycle is used to place a value into CONFADD that specifies
the PCI Bus, the device on that bus, the function within the device, and a specific configuration register of the
device function being accessed. CONFADD[31] must be 1 to enable a configuration cycle. Then, CONFDATA
becomes a window onto four bytes of configuration space specified by the contents of CONFADD. Read/write
accesses to CONFDATA generates a PCI configuration cycle to the address specified by CONFADD.
Type 0 Access: If the Bus Number field of CONFADD is 0, a type 0 configuration cycle is generated on PCI.
CONFADD[10:2] is mapped directly to AD[10:2]. The Device Number field of CONFADD is decoded onto
AD[31:11]. The TVX is Device #0 and does not pass its configuration cycles to PCI. Thus, AD11 is never
asserted. (For accesses to device #1, AD12 is asserted, etc., to Device #20 which asserts AD31.) Only one AD
line is asserted at a time. All device numbers higher than 20 cause a type 0 configuration access with no IDSEL
asserted, which results in a master abort.
Type 1 Access: If the Bus Number field of CONFADD is non-zero, a type 1 configuration cycle is generated on
PCI. CONFADD[23:2] are mapped directly to AD[23:2]. AD[1:0] are driven to 01 to indicate a Type 1
Configuration cycle. All other lines are driven to 0.
Table
2
. TVX Configuration Space
Address
Mnemonic
Register Name
Access
00−01h VID
Vendor Identification
RO
02−03h DID
Device Identification
RO
04−05h PCICMD
Command Register
R/W
06−07h PCISTS
Status Register
RO, /WC
08h RID
Revision Identification
RO
09−0Bh CLASSC
Class Code
RO
0Ch Reserved
0Dh MLT
Master Latency Timer
R/W
0Eh HEDT
Header Type
RO
0Fh BIST
BIST Register
R/W
10−4Eh Reserved
4Fh ACON
Arbitration Control
R/W
50h PCON
PCI Control
R/W
51h Reserved
52h CC
Cache Control
R/W
53h CCE
Cache Control Extended
R/W
82437VX (TVX) AND 82438VX (TDX) E
22
PRELIMINARY
Table
2
. TVX Configuration Space
Address
Mnemonic
Register Name
Access
54−55h SDRAMC
SDRAM Control
R/W
55h Reserved
56h DRAMEC
DRAM Extended Control
R/W
57h DRAMC
DRAM Control
R/W
58h DRAMT
DRAM Timing
R/W
59−5Fh
PAM[6:0]
Programmable Attribute Map (7 registers)
R/W
60−64h
DRB[3:0]
DRAM Row Boundary (5 registers)
R/W
65−66h Reserved
67h DRTH
DRAM Row Type High
R/W
68h DRT
DRAM Row Type Low
R/W
69h TRDT
PCI TRDY Timer
R/W
70h MTT
Multi-Transaction Timer
R/W
71h Reserved
72h SMRAM
System Management RAM Control
R/W
73h SMBCR
Shared Memory Buffer Control
R/W
74h SMBSA
Shared Memory Buffer Start Address
R/W
76−77h Reserved
78h GCRLT
Graphics Controller Latency Timers
R/W
79−FFh Reserved
3.2.1. VIDVENDOR IDENTIFICATION REGISTER
Address Offset: 00−01h
Default Value: 8086h
Attribute: Read Only
The VID Register contains the vendor identification number. This 16-bit register combined with the Device
Identification Register uniquely identify any PCI device. Writes to this register have no effect.
Bit
Description
15:0
Vendor Identification Number. This is a 16-bit value assigned to Intel.
E82437VX (TVX) AND 82438VX (TDX)
23
PRELIMINARY
3.2.2. DIDDEVICE IDENTIFICATION REGISTER
Address Offset: 02−03h
Default Value: 7030h
Attribute: Read Only
This 16-bit register combined with the Vendor Identification register uniquely identifies any PCI device. Writes to
this register have no effect.
Bit
Description
15:0
Device Identification Number. This is a 16-bit value assigned to the TVX.
3.2.3. PCICMDPCI COMMAND REGISTER
Address Offset: 04−05h
Default: 0006h
Access: Read/Write
This 16-bit register provides basic control over the TVX's ability to respond to PCI cycles. The PCICMD Register
in the TVX enables and disables the assertion of SERR# and PCI master accesses to main memory.
Bit
Descriptions
15:10
Reserved.
9
Fast Back-to-Back (FB2B). (Not Implemented) This bit is hardwired to 0.
8
SERR# Enable (SERRE). (Not Implemented) This bit is hardwired to 0.
7
Address/Data Stepping. (Not Implemented) This bit is hardwired to 0.
6
Parity Error Enable (PERRE). (Not Implemented) This bit is hardwired to 0.
5
Video Pallet Snooping (VPS). (Not Implemented) This bit is hardwired to 0.
4
Memory Write and Invalidate Enable (MWIE). (Not Implemented) This bit is hardwired to 0.
The TVX will never use the Memory Write and Invalidate PCI command.
3
Special Cycle Enable (SCE). (Not Implemented) This bit is hardwired to 0, as the TVX does
not respond to PCI special cycles.
2
Bus Master Enable (BME). (Not Implemented) The TVX does not support disabling of its bus
master capability on the PCI Bus. This bit is hardwired to 1.
1
Memory Access Enable (MAE). 1=Enable PCI master access to main memory, if the PCI
address selects enabled DRAM space; 0=Disable (TVX does not respond to main memory
accesses).
0
I/O Access Enable (IOAE). (Not Implemented) This bit is hardwired to 0. The TVX does not
respond to PCI I/O cycles.
82437VX (TVX) AND 82438VX (TDX) E
24
PRELIMINARY
3.2.4. PCISTSPCI STATUS REGISTER
Address Offset: 06−07h
Default Value: 0200h
Access: Read Only, Read/Write Clear
PCISTS is a 16-bit status register that reports the occurrence of a PCI master abort and PCI target abort.
PCISTS also indicates the DEVSEL# timing that has been set by the TVX hardware.
Bit
Descriptions
15
Detected Parity Error (DPE). (Not Implemented) This bit is hardwired to 1.
14
Signaled System Error (SSE). (Not Implemented) This bit is hardwired to 0.
13
Received Master Abort Status (RMAS)
R/WC. When the TVX terminates a Host-to-PCI
transaction (TVX is a PCI master) with an unexpected master abort, this bit is set to 1. Note that
master abort is the normal and expected termination of PCI special cycles. Software resets this bit
to 0 by writing a 1 to it.
12
Received Target Abort Status (RTAS)
R/WC. When a TVX-initiated PCI transaction is
terminated with a target abort, RTAS is set to 1. Software resets RTAS to 0 by writing a 1 to it.
11
Signaled Target Abort Status (STAS). (Not Implemented) This bit is hardwired to 0. The TVX
never terminates a PCI cycle with a target abort.
10:9
DEVSEL# Timing (DEVT)
RO. This 2-bit field indicates the timing of the DEVSEL# signal when
the TVX responds as a target, and is hard-wired to the value 01b (medium) to indicate the slowest
time that DEVSEL# is generated.
8
Data Parity Detected (DPD). (Not Implemented) This bit is hardwired to 0.
7
Fast Back-to-Back (FB2B). (Not Implemented) This bit is hardwired to 0.
6
User Defined Format (UDF). (Not Implemented) This bit is hardwired to 0.
5
66 MHz PCI Capable (66C). (Not Implemented) This bit is hardwired to 0.
4:0
Reserved.
3.2.5. RIDREVISION IDENTIFICATION REGISTER
Address Offset: 08h
Default Value: 00h (A0 stepping)
Access: Read Only
This register contains the revision number of the TVX. These bits are read only and writes to this register have
no effect. For the A-0 Stepping, this value is 00h.
Bit
Description
7:0
Revision Identification Number. This is an 8-bit value that indicates the revision identification
number for the TVX.
E82437VX (TVX) AND 82438VX (TDX)
25
PRELIMINARY
3.2.6. CLASSCCLASS CODE REGISTER
Address Offset: 09−0Bh
Default Value: 060000h
Access: Read Only
This register contains the device programming interface information related to the Sub-Class code and Base-
Class code definition for the TVX. This register also identifies the Base-Class code and the function Sub-Class in
relation to the Base-Class code.
Bit
Description
23:16
Base-Class Code (BCC). 06=Bridge device.
15:8
Sub-Class Code (SCC). 00h=Host bridge.
7:0
Programming Interface (PI). 00h = Hardwired as a host-to-PCI bridge.
3.2.7. HEDTHEADER TYPE REGISTER
Address Offset: 0Eh
Default Value: 00h
Access: Read Only
The Header Type Register identifies the TVX as a single-function device.
Bit
Description
7:0
Device Type (DEVICET). 00h=Single-function PCI device.
3.2.8. MLTMASTER LATENCY TIMER REGISTER
Address Offset: 0Dh
Default Value: 00h
Access: Read/Write
MLT is an 8-bit register that controls the amount of time the TVX, as a bus master, can burst data on the PCI
Bus. The Count Value is an 8-bit quantity. However, MLT[2:0] are hardwired to 0. MLT is used to guarantee the
host CPU a minimum amount of the system resources as described in the PCI Bus Arbitration section.
Bit
Description
7:3
Master Latency Timer Count Value. The number of clocks programmed in the MLT represents
the guaranteed time slice (measured in PCI clocks) allotted to the TVX, after which it must
surrender the bus as soon as other PCI masters request the bus. The default value of MLT is 00h
or 0 PCI clocks.
2:0
Reserved. Hardwired to 0.
82437VX (TVX) AND 82438VX (TDX) E
26
PRELIMINARY
3.2.9. BISTBIST REGISTER
Address Offset: 0Fh
Default Value: 00h
Access: Read/Write
The Built In Self Test (BIST) function is not supported by the TVX. Writes to this register have no affect.
Bit
Descriptions
7
BIST Supported
RO. 0=Disable the BIST function.
6
Start BIST. This function is not supported and writes have no affect.
5:4
Reserved.
3:0
Completion Code
RO. This field always returns 0 when read and writes have no affect.
3.2.10. ACONARBITRATION CONTROL REGISTER
Address Offset: 4Fh
Default Value: 00h
Access: Read/Write
The ACON Register enables and disables features related to PCI arbitration and PCI 2.1 compliance.
Bit
Description
7
Extended CPU-to-PIIX PHLDA# Signaling Enable (XPLDE). When XPLDE=1, the TVX adds
the following additional signalling to singal PHLDA# (i.e., in addition to the normal CPU/PIIX
PHOLD/PHLDA# protocol):
1. When the TVX begins a PCI read/write transaction, it asserts PHLDA# for 1 PCLK within the
address phase of the transaction.
2. If the CPU is attemp[ting a LOCKed cycle and lock has been established (i.e., PLOCK# was
negated in the address phase), PHLDA# remains asserted for one additional clock following
the address phase.
6:4
Reserved.
3
CPU Priority Enable (CPE). 1=CPU gets PCI Bus after two PCI slots (SMBA design). 0=three
PCI slots (Non-SMBA design).
2:0
Reserved.
E82437VX (TVX) AND 82438VX (TDX)
27
PRELIMINARY
3.2.11. PCONPCI CONTROL
Address Offset: 50h
Default Value: 00h
Access: Read/Write
The PCON Register enables/disables features related to PCI Bus that is not already covered in the required PCI
configuration space.
Bit
Description
7:4
Reserved.
3
PCI Concurrency Enable (PCE). 1 = CPU can access DRAM and L2 while a non-PIIX3 PCI
master is targeting peer PCI devices. 0 (default) = CPU is held off of the bus during all PCI
master cycles. This bit should be set to 1 by the BIOS during normal operation.
2:0
Reserved.
3.2.12. CCCACHE CONTROL REGISTER
Address Offset: 52h
Default Value: SSSS0010 (S = Strapping option)
Access: Read/Write
This 8-bit register defines the secondary cache operations. The CC Register enables/disables the second level
cache, adjusts cache size, selects the cache write policy, selects the caching policy when CACHE# is negated
on reads, informs the TVX how the SRAMs are connected, and defines the cache SRAM type. After a hard
reset, CC[7:4] reflect the signal levels on the Host address lines A[31:28].
Bit
Description
7:6
Secondary Cache Size (SCS). This field reflects the inverted signal level on the A[31:30] pins at
the rising edge of the CPURST signal. The default values can be overwritten with subsequent
writes to the CC Register. The options are:
Bits[7:6] Secondary Cache Size
0 0 Cache not populated
0 1 256 Kbytes
1 0 512 Kbytes
1 1 Reserved
1. When SCS=00, the secondary cache is disabled.
2. To enable the L2 cache, SCS must be non-zero and the FLCE bit must be 1. ( L2 cache
cannot be enabled unless the L1 cache is also enabled)
82437VX (TVX) AND 82438VX (TDX) E
28
PRELIMINARY
Bit
Description
5:4
L2 RAM Type (L2RAMT). This field reflects the inverted signal level on the A[29:28] pins at the
rising edge of the CPURST signal. The reset values can be overwritten with subsequent writes to
the CC Register. The options are:
Bits[5:4] SRAM Type
0 0 Pipelined Burst SRAM or DRAM Cache
0 1 Reserved
1 0 Async SRAM
1 1
Two banks of Pipelined Burst SRAM or DRAM Cache
NOTE
When 512K Pipelined Burst SRAM L2 mode is selected (via SCS and L2RAMT), NA# is
delayed one HCLK during burst reads from L2 to ensure that the active bank is not deselected
too early by pipelining a cycle to the opposite bank. It is an illegal combination to use
asynchronous SRAMs with SDRAMs. BIOS should ensure this combination is not selected
when programming this register.
3
NA Disable (NAD). When NAD=1, the TVX’s NA# pin is never asserted. When NAD=0, NA#
assertion is dependent on the cache type and size selected (via SRAMT[SCS]). Note that NAD
must be set to 1 if the NA pin of the TVX is not connected to the processor. This bit should be reset
to 0 for normal operation in systems that connect NA# to the processor.
2
Reserved.
1
Secondary Cache Force Miss or Invalidate (SCFMI). When SCFMI=1, the L2 hit/miss detection
is disabled, and all tag lookups result in a miss. If the L2 is enabled, the cycle is processed as a
miss. If the L2 is populated but disabled (FLCE=0) and SCFMI=1, any CPU read cycle invalidates
the selected tag entry. When SCFMI=0, normal L2 cache hit/miss detection and cycle processing
occurs.
Software can flush the cache (cause all modified lines to be written back to main memory) by
setting SCFMI to 1 with the L2 cache enabled (SCS
≠
00 and FLCE=1), and reading all L2 cache tag
address locations.
0
First Level Cache Enable (FLCE): FLCE enables/disables the first level cache. When FLCE=1,
the TVX responds to CPU cycles with KEN# asserted for cacheable memory cycles. When
FLCE=0, KEN# is always negated and line fills to either the first level or L2 cache are prevented.
Note that, when FLCE=1 and SCFMI=1, writes to the cache are also forced as misses. Thus, it is
possible to create incoherent data between main memory and the L2 cache. A summary of
FLCE/SCFMI bit interactions is as follows:
FLCE SCFMI
L2 Cache Result
0 0 Disabled
0 1 Disabled; tag invalidate on reads
1 0 Normal L2 cache operation (dependent on SCS)
1 1
Enabled; miss forced on reads/writes
E82437VX (TVX) AND 82438VX (TDX)
29
PRELIMINARY
3.2.13. CCECACHE CONTROL EXTENDED REGISTER
Address Offset: 53h
Default Value: 14h
Access: Read/Write (bit 5 is read only)
This register defines the refresh rate, in HCLKs, for the DRAM Cache implementation. The register also provides
a read only bit for DRAM Cache presence detection. The value of this bit directly reflects the strapping on the
KRQAK pin, as determined at reset.
Bit
Description
7:6
Reserved.
5
DRAM Cache Detect (DCD)
RO. 1 = DRAM Cache present. 0=DRAM Cache not present. This
read only bit directly reflects the strapping on the KRQAK pin, as determined at reset.
4:0
DRAM Cache Refresh Timer (DCRT)
R/W. These bits determine the time TVX remains idle, in
HCLks, during a DRAM cache refresh sequence. The default value sets the refresh timer to 20
HCLKs.
82437VX (TVX) AND 82438VX (TDX) E
30
PRELIMINARY
3.2.14. SDRAMCSDRAM CONTROL REGISTER
Address Offset: 54−55h
Default Value: 0000h
Access: Read/Write
This register controls the SDRAM mode, CAS# latency, RAS# timing, and SDRAM Turbo enable/disable.
Bit
Description
15:9
Reserved.
8:6
Special SDRAM Mode Select (SSMS). This field selects 1 of 4 special SDRAM modes used for
testing and initialization. The NOP command must be programmed first before any other command
can be issued. After the DRAM detection process has completed, bits [8:6] must remain at
“
000
”
during normal DRAM operation.
Bits [8:6] Mode
000 Normal SDRAM mode (default).
001 NOP Command Enable (NOPCE). Forces all CPU cycles to DRAM to generate a
SDRAM NOP command on the memory interface.
010 All Banks Prechage Command Enable (ABPCE). CPU cycles to DRAM are
converted to an all banks precharge command on the memory interface.
011 Mode Register Command Enable (MRCE). CPU cycles to DRAM are converted to
MRS commands to the memory interface. The command is driven on the MA[11:0] lines:
MA[2:0]=010 for burst of 4 mode, MA3=1 for interleave wrap type mode. MA4=the value
in the CAS# Latency bit, MA[6:5]=01, and MA[11:7]=00000.
BIOS selects an appropriate host address for each Row of memory such that the right
commands are generated on the Memory Address MA[11:0] lines. BIOS needs to be
cognizant of the mapping of the host addresses to memory addresses (e.g., a
hostaddress of 1D0h sets up the Mode registers in Row 0 of SDRAM with Burst length
of 4, Wrap type of interleaved. and CAS latency of 3).
100 CBR Cycle Enable(CBRC). 100= CPU cycles to DRAM are converted to SDRAM CBR
refresh cycles on the memory interface.
101 Reserved
11X Reserved
5
Reserved.
4
CAS# Latency (CL). 1=CAS# latency of 2 for all SDRAM cycles. 0=CAS# latency of 3. RAS# to
CAS# delay is also controlled by this bit. When programmed for 2 clock CAS# latency, a RAS# to
CAS# delay of 2 HCLKs is provided. When programmed for 3 clock CAS# latency, a RAS# to CAS#
delay of 3 HCLKs is provided.
3
RAS# Timing (RT). This bit controls RAS# precharge, RAS# active to precharge time and Refresh
to RAS# active delay (in HCLKs):
Bit3 RAS# RAS# active Refresh to
precharge to precharge RAS# active
0 3 5 8
1 3 4 7
2:0
Reserved.
E82437VX (TVX) AND 82438VX (TDX)
31
PRELIMINARY
The following table lists the CAS# latency, RAS# precharge, RAS# active to precharge, and refresh to RAS#
active requirements for the various speed grades.
Table
3
. SDRAM Timing Values vs Speed Upgrades
Speed Grade
(MHz)
CAS# Latency
HLCKs
Frequency
(Mhz)
RAS# Precharge
Required (Trp)
RAS# Active to
Precharge
Required (Tras)
Refresh to RAS#
Active Required
(Trc)
66.67 MHz
3
50/60/66 MHz
3 HCLKs
5 HCLKs
8 HCLKs
66.67 MHz
2
50/60/66 MHz
3 HCLKs
5 HCLKs
8 HCLKs
50 MHz
2
50 MHz
3 HCLKs
4 HCLKs
7 HCLKs
3.2.15. DRAMECDRAM EXTENDED CONTROL REGISTER
Address Offset: 56h
Default Value: 52h
Access: Read/Write
This 8-bit register contains additional controls for main memory DRAM operating modes and features.
Bit
Description
7
Reserved.
6
Refresh RAS# Assertion (RRA). 1=5 clocks (default). 0=4 clocks. This bit controls the number of
clocks RAS# is asserted for Refresh cycles.
5
Fast EDO Path Select(FEPS). When FEPS=1, a fast path is selected for CPU-to-DRAM read
cycles for the leadoff. This is valid for EDO DRAMs only in both a cache and a cacheless system.
The fast path is selected by assertion of MXS pin. This result in a 1 HCLK pull-in for all read leadoff
latencies for EDO DRAMs. (i.e., Page hits, Page misses and Row Misses).
The selection of the Fast Path for the burst cycles depends on the value of DRAM Read Burst
Timing (DRBT). The Fast Path is selected for the Burst cycles only if DRBT =10 (x222). The slow
path is selected for the burst cycles if DRBT = 11 (x322).
The FEPS bit should not be set to 1 by BIOS if the EDO Burst rate is either x333, x444, or
Asynchronous cache is in the system.
4
Speculative Leadoff Disable (SLD). 1= Speculative leadoff disabled (default) . 0= Speculative
leadoff enabled. This bit defaults to 1 (disabled), and should be set to 0 in a system without an L2
cache. In a system with an L2 cache, this bit should be set to 1.
3
Reserved.
2:1
Memory Address Drive Strength (MAD). This field controls the strength of the output buffers
driving the MA pins. If one or two rows of memory are populated, set this field to 01 (10 mA).
Otherwise, set the field to 10 (16 mA).
Bit [2:1] MA[11:0]
00 Reserved
01 10 mA (default)
10 16 mA
11 Reserved
0
DRAM Symmetry Detect Mode (DSDM). If set to 1, this bit forces some of the MA lines to a fixed
value to allow software to detect DRAM symmetry type on a row by row basis.
82437VX (TVX) AND 82438VX (TDX) E
32
PRELIMINARY
3.2.16. DRAMCDRAM CONTROL REGISTER
Address Offset: 57h
Default Value: 01h
Access: Read/Write
This 8-bit register controls main memory DRAM operating modes and features.
Bit
Description
7:6
Hole Enable (HEN). This field enables a memory hole in DRAM space. CPU cycles matching an
enabled hole are passed on to PCI. PCI cycles matching an enabled hole are ignored by the TVX (no
DEVSEL#). Note that a selected hole is not remapped. Also, note that this hole enable overrides the
SMBA address range if any overlap occurs. (i.e., The hole will exist within the SMBA range).
Bits[7:6] Hole Enabled
00 None
01 512 Kbytes
−
640 Kbytes
10 15 Mbytes
−
16 Mbytes
11
14 Mbytes
−
16 Mbytes
5:4
Reserved.
3
EDO Detect Mode Enable (EDME). This bit, if set to a 1, enables a special timing mode for BIOS to
detect EDO DRAM type on a bank-by-bank basis. Once all DRAM row banks have been tested for
EDO, the EDME bit should be set to 0. Otherwise, performance will be seriously impacted.
2:0
DRAM Refresh Rate (DRR). The DRAM refresh rate is adjusted according to the frequency
selected by this field. Note that refresh is also disabled via this field, and that disabling refresh results
in the eventual loss of DRAM data.
Bits[2:0] Host Bus Frequency
000 Refresh Disabled
001 50 MHz
010 60 MHz
011 66 MHz
1XX Reserved
DRR(0) is initialized to the inverted signal level on A27 at the rising edge of reset. Since the A27 pin
contains an internal weak pull-down, unless an external resistor exists, the field will be initialized
identical to the 82430FX PCIset. Subsequent writes to this field will override the reset strap value.
BIOS can use the value in DRR[0] to determine if the system is 60 MHz (external pullup) or 66 MHz
(no strapping).
E82437VX (TVX) AND 82438VX (TDX)
33
PRELIMINARY
3.2.17. DRAMTDRAM TIMING REGISTER
Address Offset: 58h
Default Value: 00h
Access: Read/Write
This 8-bit register controls main memory DRAM timings. For SDRAM specific timing control, see the SDRAMC
register.
Bit
Description
7
Fast MA to RAS# Delay (FMRD). 1=1 clock (MA setup to RAS# assertion). 0=2 clocks. The
DRAM Row Miss timings are controlled by FMRD. Note that FMRD timing adjustments are
independent of DLT timing adjustment.
6:5
DRAM Read Burst Timing (DRBT). The DRAM read burst timings are controlled by the DRBT
field. Slower rates may be required in certain system designs to support layouts with longer trace
lengths or slower memories. Most system designs will be able to use one of the faster burst mode
timings. The timing used depends on the type of DRAM on a per-bank basis, as indicated by the
DRT register.
The x322 timings for EDO burst rate should be used only when FEPS=1 (DRAMEC register) and
the timings for the EDO Burst rate for X222 have a negative margin. This forces the MXS to be
negated after the leadoff, thus, selecting the fast path for the leadoff and the slow path for the Burst
cycles.
When FEPS=1 and the EDO Burst rate is set to x222, the EDO read cycle is selected through the
fast path for both leadoff and the burst cycles. When FEPS=0 and the EDO Burst rate is set to
x222, the EDO read cycle is selected through the slow path for both leadoff and the burst cycles.
DRBT EDO Burst Rate SPM Burst Rate
00 x444 x444
01 x333 x444
10 x222 x333
11 x322 x333
NOTE
For systems with five rows of memory, at 66 MHz with 60 ns memory or 60 MHz with 70 ns
memory, the DRBT field must be set to 01.
4:3
DRAM Write Burst Timing (DWBT). The DRAM write burst timings are controlled by the DWBT
field. Slower rates may be required in certain system designs to support layouts with longer trace
lengths or slower memories. Most system designs will be able to use one of the faster burst mode
timings.
DWBT EDO/SPM Burst Rate
00 x444
01 x333
10 x222
11 Reserved
2
Fast RAS to CAS Delay (FRCD). 1=2 clocks (RAS active to CAS active delay). 0=3 clocks. The
DRAM row and page miss leadoff timings are controlled by the FRCD bit. The slower timing may
be required in certain system designs to support layouts with longer trace lengths or slower
memories. Note that FRCD timing adjustments are independent to DLT timing adjustments.
82437VX (TVX) AND 82438VX (TDX) E
34
PRELIMINARY
Bit
Description
1:0
DRAM Leadoff Timing (DLT). The DRAM leadoff timings are controlled by the DLT bits. Slower
leadoffs may be required in certain system designs to support layouts with longer trace lengths or
slower memories. The row miss leadoff timings are summarized below for EDO/SPM reads and
writes.
Changing DLT effects the Row Miss and Page Miss timings only (e.g., DLT=01 is one clock faster
than DLT=00 on Row Miss and Page Miss timings). These bits control MA setup to CAS#
assertion.
DLT does not effect page hit timings (e.g., DLT=00 or DLT=01 have same page hit timings for
reads and writes; e.g., for reads it would be (10-3 = 7) clocks for DLT=00 or DLT=01).
DLT
Read Leadoff Write Leadoff RAS# Precharge
00 11 7 3
01 10 6 3
10 11 7 4
11 10 6 4
FRCD, SLE, FEPS and FMRD bits have cumulative effect on the leadoff timings. The above
leadoff represent timings with FRCD=0, SLE=0, FEPS=0 and FMRD = 0. Refer to the
“
EDO/SPM
leadoff timing calculation
”
table under DRAM Performance Section to see the effect of these bits on
the leadoff.
3.2.18. PAMPROGRAMMABLE ATTRIBUTE MAP REGISTERS (PAM[6:0])
Address Offset: PAM0 (59h) PAM6 (5Fh)
Default Value: 00h
Attribute: Read/Write
The TVX allows programmable memory and cacheability attributes on 14 memory segments of various sizes in
the 640-Kbyteto 1-Mbyte address range. Seven Programmable Attribute Map (PAM) Registers are used to
support these features. Three bits are used to specify L1 cacheability and memory attributes for each memory
segment. These attributes are:
RE Read Enable. When RE=1, the CPU read accesses to the corresponding memory segment are
directed to main memory. Conversely, when RE=0, the CPU read accesses are directed to PCI.
WE Write Enable. When WE=1, the CPU write accesses to the corresponding memory segment are
directed to main memory. Conversely, when WE=0, the CPU write accesses are directed to PCI.
CE Cache Enable. When CE=1, the corresponding memory segment is L1 cacheable. CE must not be set
to 1 when RE is reset to 0 for any particular memory segment. When CE=1 and WE=0, the
corresponding memory segment is cached in the first level cache only on CPU code read cycles.
The RE and WE attributes permit a memory segment to be Read Only, Write Only, Read/Write, or disabled. For
example, if a memory segment has RE=1 and WE=0, the segment is Read Only. The characteristics for memory
segments with these read/write attributes are described in Table 4.
E82437VX (TVX) AND 82438VX (TDX)
35
PRELIMINARY
Table
4
. Attribute Definition
Read/Write
Attribute
Definition
Read Only
Read cycles: CPU cycles are serviced by the DRAM in a normal manner.
Write cycles: CPU initiated write cycles are ignored by the DRAM interface as well as the
cache. Instead, the cycles are passed to PCI for termination.
Areas marked as Read Only are L1 cacheable for Code accesses only. These regions are
not cached in the second level cache.
Write Only
Read cycles: All read cycles are ignored by the DRAM interface as well as the second level
cache. CPU-initiated read cycles are passed onto PCI for termination. The write only state
can be used while copying the contents of a ROM, accessible on PCI, to main memory for
shadowing, as in the case of BIOS shadowing.
Write cycles: CPU write cycles are serviced by the DRAM and L2 cache in a normal manner.
Read/Write
This is the normal operating mode of main memory. Both read and write cycles from the CPU
and PCI are serviced by the DRAM and L2 cache interface.
Disabled
All read and write cycles to this area are ignored by the DRAM and cache interface. These
cycles are forwarded to PCI for termination.
Each PAM Register controls two regions, typically 16 Kbytes in size. Each of these regions has a 4-bit field. The
four bits that control each region have the same encoding and are defined in Table 5.
PCI master access to DRAM space is also controlled by the PAM Registers. When the PAM programming
indicates a region is writeable, PCI master writes are accepted (DEVSEL# generated). When the PAM
programming indicates a region is readable, PCI master reads are accepted. When a PCI write to a non-
writeable DRAM region, or a PCI read to a non-readable DRAM region is seen, the TVX does not accept the
cycle (DEVSEL# is not asserted). PCI master accesses to enable memory hole regions are not accepted.
Table
5
. Attribute Bit Assignment
Bits [7, 3]
Reserved
Bits [6, 2]
Cache Enable
Bits [5, 1]
Write Enable
Bits [4, 0]
Read Enable
Description
x x 0 0
DRAM disabled, accesses directed
to PCI
x0 0 1
read only, DRAM write protected,
non-cacheable
x1 0 1
read only, DRAM write protected, L1
cacheable for code accesses only
x0 1 0
write only
x0 1 1
read/write, non-cacheable
x1 1 1
read/write, cacheable
82437VX (TVX) AND 82438VX (TDX) E
36
PRELIMINARY
As an example, consider a BIOS that is implemented on the expansion bus. During the initialization process the
BIOS can be shadowed in main memory to increase the system performance. When a BIOS is shadowed in
main memory, it should be copied to the same address location. To shadow the BIOS, the attributes for that
address range should be set to write only. The BIOS is shadowed by first doing a read of that address. This read
is forwarded to the expansion bus. The CPU then does a write of the same address, which is directed to main
memory. After the BIOS is shadowed, the attributes for that memory area are set to read only so that all writes
are forwarded to the expansion bus.
Table
6
. PAM Registers and Associated Memory Segments
PAM Reg
Attribute Bits
Memory Segment
Comments
Offset
PAM0[3:0]
Reserved 59h
PAM0[7:4]
RCE WE RE 0F0000−0FFFFFh
BIOS Area
59h
PAM1[3:0]
RCE WE RE 0C0000−0C3FFFh
ISA Add-on BIOS
5Ah
PAM1[7:4]
RCE WE RE 0C4000−0C7FFFh
ISA Add-on BIOS
5Ah
PAM2[3:0]
RCE WE RE 0C8000−0CBFFFh
ISA Add-on BIOS
5Bh
PAM2[7:4]
RCE WE RE 0CC000−0CFFFFh
ISA Add-on BIOS
5Bh
PAM3[3:0]
RCE WE RE 0D0000−0D3FFFh
ISA Add-on BIOS
5Ch
PAM3[7:4]
RCE WE RE 0D4000−0D7FFFh
ISA Add-on BIOS
5Ch
PAM4[3:0]
RCE WE RE 0D8000−0DBFFFh
ISA Add-on BIOS
5Dh
PAM4[7:4]
RCE WE RE 0DC000−0DFFFFh
ISA Add-on BIOS
5Dh
PAM5[3:0]
RCE WE RE 0E0000−0E3FFFh
BIOS Extension
5Eh
PAM5[7:4]
RCE WE RE 0E4000−0E7FFFh
BIOS Extension
5Eh
PAM6[3:0]
RCE WE RE 0E8000−0EBFFFh
BIOS Extension
5Fh
PAM6[7:4]
RCE WE RE 0EC000−0EFFFFh
BIOS Extension
5Fh
NOTE:
The CE bit should not be changed while the L2 cache is enabled.
DOS Application Area (00000−−9FFFh)
Read, write, and cacheability attributes are always enabled and are not programmable for the 0 - 640-Kbyte
DOS application region.
Video Buffer Area (A0000−−BFFFFh)
This 128-Kbyte area is not controlled by attribute bits. CPU-initiated cycles in this region are always forwarded to
PCI for termination. This area is not cacheable.
Expansion Area (C0000−−DFFFFh)
This 128-Kbyte area is divided into eight 16-Kbyte segments. Each segment can be assigned one of four
Read/Write states: read-only, write-only, read/write, or disabled. Memory that is disabled is not remapped.
Cacheability status can also be specified for each segment.
Extended System BIOS Area (E0000−−EFFFFh)
This 64-Kbyte area is divided into four 16-Kbyte segments. Each segment can be assigned independent
cacheability, read, and write attributes. Memory segments that are disabled are not remapped elsewhere.
E82437VX (TVX) AND 82438VX (TDX)
37
PRELIMINARY
System BIOS Area (F0000−−FFFFFh)
This area is a single 64-Kbyte segment. This segment can be assigned cacheability, read, and write attributes.
When disabled, this segment is not remapped.
Extended Memory Area (100000−−FFFFFFFFh)
The extended memory area can be split into several parts;
•Flash BIOS area from 4 Gbytes to 4 Gbytes minus 512 Kbytes (aliased on ISA at 16 Mbytes minus 15.5
Mbytes)
•DRAM Memory from 1 Mbyte to a maximum of 512 Mbytes
•PCI Memory space from the top of DRAM to 4 Gbytes minus 512 Kbytes
On power-up or reset the CPU vectors to the Flash BIOS area, mapped in the range of 4 Gbytes to 4 Gbytes
minus 512 Kbytes. This area is physically mapped on the expansion bus. Since these addresses are in the
upper 4-Gbyte range, the request is directed to PCI.
The DRAM memory space can occupy extended memory from a minimum of 1 Mbyte up to 512 Mbytes. This
memory is cacheable. PCI memory space from the top of main memory to 4 Gbytes is always non-cacheable.
3.2.19. DRBDRAM ROW BOUNDARY REGISTERS
Address Offset: 60h (DRB0)64h (DRB4)
Default Value: 02h
Access: Read/Write
The TVX supports 5 rows of DRAM. Each row is 64 bits wide. The DRAM Row Boundary Registers define upper
and lower addresses for each DRAM row. Contents of these 8-bit registers represent the boundary addresses in
4-Mbyte granularity. If the 5th row is implemented, it must be fixed at 8MB (using 1Mx16 devices), in a SMBA
system and can be set to either 8MB (using 1Mx16 deivces) or 16MB (using 2Mx8 devices), in a non-SMBA
system. Refer to the 5th row discussion in the DRAM section for addtional information on 5th Row support.
DRB0 = Total amount of memory in row 0 (in 4 Mbytes)
DRB1 = Total amount of memory in row 0 + row 1 (in 4 Mbytes)
DRB2 = Total amount of memory in row 0 + row 1 + row 2 (in 4 Mbytes)
DRB3 = Total amount of memory in row 0 + row 1 + row 2 + row 3 (in 4 Mbytes)
DRB4 = Total amount of memory in row 0 + row 1 + row 2 + row 3 + row 4 (in 4 Mbytes)
The DRAM array can be configured with 512kx32, 1Mx32, 2Mx32, and 4Mx32, SIMMs. Each register defines an
address range that causes a particular RAS# line to be asserted (e.g. if the first DRAM row is 8 Mbytes in size,
then accesses within the 0 to 8-Mbyte range causes RAS0# to be asserted).
Bit
Description
7:6
Reserved.
5:0
Row Boundary Address. This 6-bit value is compared against address lines A[27:22] to
determine the upper address limit of a particular row. (i.e., DRB minus previous DRB = row size).
Row Boundary Address
These 6-bit values represent the upper address limits of the 4 rows (i.e., this row minus previous row = row
size). Unpopulated rows have a value equal to the previous row (row size = 0). DRB4 reflects the maximum
amount of DRAM in the system. The top of memory is determined by the value written into DRB4. DRB4 should
not be programmed to a value greater than 128 Mbytes.
Note that the total memory supported is limited to 128 Mbytes even though it is possible to populate the 5 rows
with up to 160 Mbyte. This limit must be ensured by the system BIOS. If the system detects more than 128
82437VX (TVX) AND 82438VX (TDX) E
38
PRELIMINARY
Mbytes in a SMBA system, then it must program one of the DRB’s [3:0], except DRB4, with a lesser value so
that the total system memory adds up to 128 Mbyte. In a non-SMBA system (5th row fixed at 8 Mbyte), memory
may be trimmed from any DRAM row in order to bring the total to 128 Mbytes or less. When possible, BIOS
should take away from the slowest DRAM row in a 5 row mixed FPM/EDO system. In a non-SMBA system (5th
row fixed at 16 Mbytes), the BIOS must disable the 5th row of memory.
NOTE
The Intel 430VX PCIset only supports the 5th row fixed at 8 Mbyte, in a SMBA system.
Example #1 (memory exceeding 128 Mbytes)
Five DRAM row ,SMBA system: The first 4 rows have 32 Mbytes each and the 5th row has 8 Mbytes of memory
for a total of 136 Mbytes. BIOS must program one of the first four Rows with 24 Mbytes to ensure that the 5th
row is fully accessable.
Example #2 (memory exceeding 128 Mbytes)
Five DRAM row, non-SMBA system (5th row fixed at 8 Mbytes): Rows 0 to 3 are fast EDO DRAM (32 Mbytes
per row), row 4 is 8 Mbytes of FPM DRAM (slower than the EDO) for a total of 136 Mbytes. The BIOS in this
case programs DRB4 to the same value as DRB3, effectively eliminating the 8 Mbytes in row 4. BIOS reports
the memory size as 128 Mbytes only (128 Mbytes is always the maximum).
Example #3 (memory exceeding 128 Mbytes)
Five DRAM row, non-SMBA system (5th row fixed at 16 Mbytes): Rows 0 to 3 are fast EDO DRAM (32 Mbytes
per row) and row 4 is 16 Mbytes of EDO DRAM for a total of 144 Mbytes. The BIOS, in this case, programs
DRB4 to the same value as DRB3, effectively eliminating the 16 Mbytes in row 4. BIOS reports the memory size
as 128 Mbytes only (128 Mbytes is always the maximum).
As an example of a general purpose configuration where 4 physical rows are configured for either single-sided or
double-sided SIMMs, the memory array would be configured like the one shown in Figure 2. In this configuration,
the TVX drives two RAS# signals directly to the SIMM rows. If single-sided SIMMs are populated, the even
RAS# signal is used and the odd RAS# is not connected. If double-sided SIMMs are used, both RAS# signals
are used.
NOTE
In a SMBA (four RAS) system, RAS3# should be mapped to the front of SIMM 3 and 2. This mapping
allows a single-sided SIMM to be used for SIMMs 2 and 3 in an SMBA system (i.e., last DRAM row must
always be occupied in an SMBA system).
SIM M-1 Front
SIMM -1 Back
SIM M-0 Front
SIMM-0 Back
RAS 1#
RAS 0#
CAS 7#
CAS 6#
CAS 5#
CAS4#
CAS 3#
CAS2#
CAS 1#
CAS0#
DR B0
DR B1
SIM M-3 Front
SIMM -3 Back
SIM M-2 Front
SIMM-2 Back
RAS 3#
RAS 2# D R B2
DR B3
055304
Figure 2. SIMMs and Corresponding DRB Registers
E82437VX (TVX) AND 82438VX (TDX)
39
PRELIMINARY
The following 3 examples describe how the DRB Registers are programmed for cases of single-sided and
double-sided SIMMs on a motherboard having a total of four, 8-byte or eight, 4-byte SIMM sockets.
Example #1
The memory array is populated with four single-sided 1MB x 32 SIMMs (16 Mbytes of DRAM). Two SIMMs are
required for each populated row making each populated row 8 Mbytes in size.
DRB0 = 02h populated (2 SIMMs, 8 Mbytes this row)
DRB1 = 04h populated (2 SIMMs, 8 Mbytes this row)
DRB2 = 04h empty row
DRB3 = 04h empty row
DRB4 = 04h row 4 functionality not implemented (i.e. no memory in 5th row)
Example #2
As an another example, the memory array is populated with two 2-Mbyte x 32 double-sided SIMMs (one row),
and four 4-Mbyte x 32 single-sided SIMMs (two rows), yielding a total of 96 Mbytes of DRAM. The DRB registers
are programmed as follows:
DRB0 = 04h populated with 16 Mbytes, 1/2 of double-sided SIMMs
DRB1 = 08h the other 16 Mbytes of the double-sided SIMMs
DRB2 = 10h populated with 32 Mbytes, one of the sided SIMMs
DRB3 = 18h the other 32 Mbytes of single-sided SIMMs
DRB4 = 04h row 4 functionality not implemented (i.e., no memory in 5th row)
Example #3
As another example in a SMB enabled system using row 3 as the shared DRAM row, the memory array is
populated with two 2-Mbyte x 32 double-sided SIMMs (two rows), and two 4-Mbyte x 32 single-sided SIMMs
(one row), yielding a total of 64 Mbytes of DRAM. The DRB registers are programmed as follows:
DRB0 = 08h 32 MBytes of single-sided SIMMs
DRB1 = 08h empty
DRB2 = 0Ch populated with 16 Mbytes, 1/2 of double-sided SIMMs
DRB3 = 10h the other 16 Mbytes of the double-sided SIMMs (shared row)
DRB4 = 10h row 4 functionality not implemented (ie. disabled)
3.2.20. DRTHDRAM ROW TYPE REGISTER HIGH
Address Offset: 67h
Default Value: 11h
Access: Read/Write
This 8-bit register identifies the type of DRAM used in row 4 (EDO, SPM, or SDRAM), and for EDO DRAMs or
SDRAMs, should be programmed by BIOS for optimum performance. The Intel 430VX PCIset uses these bits to
determine the correct cycle timing to use before a DRAM cycle is run and also to enable Row 4 (i.e., RAS4# or
CS4# functionality muxing on TVX pins SRASB#/RAS4# or MXS/CS4#, respectively). If these bits are changed
from their default value of 11h, row 4 is assumed to be present in the system.
Bit
Description
7:5
Reserved.
4
See bits [4,0] below.
3:1
Reserved.
82437VX (TVX) AND 82438VX (TDX) E
40
PRELIMINARY
Bit
Description
4,0
DRAM Row Type (DRT). The DRT bits select the DRAM type installed in physical DRAM Row 4.
Bits [4,0] DRAM Type value defintions
0,0 SPM DRAM
0,1 EDO DRAM
1,0 SDRAM
1,1 Row 4 signalling via TVX MXS, SRASB# pins disabled
3.2.21. DRTLDRAM ROW TYPE REGISTER LOW
Address Offset: 68h
Default Value: 00h
Access: Read/Write
The DRT Register identifies the type of DRAM used in each row; EDO, SPM (standard page mode), or SDRAM
(synchronous DRAM) used in rows 0-3. This register should be programmed by BIOS for optimum performance
if EDO DRAMs or SDRAMs are used. The hardware uses these bits to determine the correct cycle timing to use
before a DRAM cycle is run.
Bit
Description
7:0
DRAM Row Type (DRT). These bits select the DRAM type installed in each physical DRAM Row.
Each one-of-four bit pairs in this register corresponds to the DRAM row identified by the
corresponding DRB Register.
DRT Bits DRAM Row
7,3 3
6,2 2
5,1 1
4,0
0
DRT DRAM Type value definitions:
0,0 SPM DRAM
0,1 EDO DRAM
1,0 SDRAM
1,1
reserved
E82437VX (TVX) AND 82438VX (TDX)
41
PRELIMINARY
3.2.22. TRDTPCI TRDY TIMER
Address Offset: 69h
Default Value: 03h
Access: Read/Write
Software programs this register to limit the number of PCI wait-states the TVX adds during the burst portion of a
PCI master read or write cycle targeted for the TVX. The TVX response time, in PCI clocks, does not exceed the
programmable count value in this register. The default value is set to meet the maximum allowable time per the
PCI 2.0 specification.
Bit
Description
7:3
Reserved.
2:0
TRDY Time-Out Value (TTOV). This field contains the TRDY time-out value for PCI master read
and write cycles targeted for the TVX.
Bits [2:0] Time-Out Value (In PCICLKs)
000 2
001 4
010 6
011 8 (default)
1xx Reserved
3.2.23. MTTMULTI-TRANSACTION TIMER REGISTER
Address Offset: 70h
Default Value: 20h
Access: Read/Write
This register controls the amount of time that the TVX’s arbiter allows a PCI initiator to perform multiple
transactions on the PCI bus. The MTT guarantees the minimum time (measured in PCLKs) that the PCI agent
retains the ownership of the PCI bus from the initial assertion of grant.
Bit
Description
7:2
MTT Count Value. The number of clocks programmed in the MTT represents the guaranteed time
slice (in PCLKs) allotted to the current agent, after which the TVX will grant the bus as soon as
another PCI agent requests the bus. The value of 00h disables this function. The count value
should be set to multiples of 4 (i.e., 2 lsbs are ignored).
1:0
Reserved. Hardwired to 0. (i.e., counter has a resolution of 4 PCLKs)
82437VX (TVX) AND 82438VX (TDX) E
42
PRELIMINARY
3.2.24. SMRAMSYSTEM MANAGEMENT RAM CONTROL REGISTER
Address Offset: 72h
Default Value: 02h
Access: Read/Write
The System Management RAM Control Register controls how accesses to this space are treated. The Open,
Close, and Lock SMRAM Space bits function only when the SMRAM enable bit is set to a 1. Also, the OPEN bit
should be reset before the LOCK bit is set.
Bit
Description
7
Reserved.
6
SMM Space Open (DOPEN). When DOPEN=1 and DLCK=0, SMM space DRAM is made visible
even when SMIACT# is negated. This is intended to help BIOS initialize SMM space. Software
should ensure that DOPEN=1 is mutually exclusive with DCLS=1. When DLCK is set to a 1,
DOPEN is reset to 0 and becomes read only.
5
SMM Space Closed (DCLS). When DCLS=1 SMM space DRAM is not accessible to data
references, even if SMIACT# is asserted. Code references may still access SMM space DRAM.
This allows SMM software to reference "through" SMM space to update the display even when
SMM space is mapped over the VGA range. Software should ensure that DOPEN = 1 is mutually
exclusive with DCLS = 1.
4
SMM Space Locked (DLCK). When DLCK is set to a 1, then DOPEN is set to 0 and both DLCK
and DOPEN become read only. DLCK can be set to 1 but can only be cleared by a power-on
reset. The combination of DLCK and DOPEN provide convenience with security. The BIOS can
use the DOPEN function to initialize SMM space and then use DLCK to "lock down" SMM space in
the future so that no application software (or BIOS itself) can violate the integrity of SMM space,
even if the program has knowledge of the DOPEN function.
3
SMRAM Enable (SMRAME). When SMRAME=1, the SMRAM function is enabled, providing 128
Kbytes of DRAM accessible at the A0000h address while in SMM (ADS# with SMIACT#).
2:0
SMM Space Base Segment (DBASESEG). This field programs the location of SMM space.
“
SMM DRAM
”
is not remapped. It is simply
“
made visible
”
if the conditions are right to access SMM
space. Otherwise, the access is forwarded to PCI. DBASESEG=010 is the only allowable setting,
which selects the SMM space as A0000-BFFFFh. All other values are reserved. PCI initiators are
not allowed access to SMM space.
NOTES:
Table 7 summarizes the operation of SMRAM space cycles targeting SMI space addresses (A-segment).
E82437VX (TVX) AND 82438VX (TDX)
43
PRELIMINARY
Table
7
. SMRAM Space Cycles
SMRAME DLCK DCLS DOPEN SMIACT#
Code Fetch
Data Reference
0X X X X PCI PCI
1 0 0 0 0 DRAM DRAM
1 0 X0 1 PCI PCI
1 0 0 1 XDRAM DRAM
1 0 1 0 0 DRAM PCI
1 0 1 1 XINVALID INVALID
1 1 0 X0DRAM DRAM
1 1 X X 1PCI PCI
1 1 1 X0DRAM PCI
3.2.25. SMBCRSMB CONTROL REGISTER
Address Offset: 73h
Default Value: 00h
Access: Read/Write
This 8-bit register controls the shared memory buffer structure.
Bit
Description
7:2
Reserved.
1
Shared Memory Buffer Enable (SMBE). When SMBE=1 and SMBR=1, the DRAM address
range defined between SMBSA and top of memory becomes the shared memory buffer area with
read, write, and non-cacheable attributes. The TVX multiplexed REQ3#/MREQ# and
GNT3#/MGNT# pins have the SMBA MREQ# and MGNT# functions, correspondingly. When
SMBE=0, the REQ3#/GNT3# pins on the TVX are treated as a fourth PCI master request pair.
0
Shared Memory Buffer Access Re-direct (SMBR). When SMBE=1 and SMBR=0 and a valid
SMBA range has been defined by the SMBSA and top of memory values (i.e., non-zero range), the
SMBA range effectively becomes a ‘hole’ in system DRAM memory. All accesses to this hole are
passed on to the PCI bus for termination. This can be used to re-direct CPU accesses to the video
frame buffer across the PCI bus into the video controller itself.
82437VX (TVX) AND 82438VX (TDX) E
44
PRELIMINARY
3.2.26. SMBSASMB START ADDRESS
Address Offset: 74h
Default Value: 0Eh
Access: Read/Write
This 8-bit register defines the system DRAM start address for the shared memory buffer structure. Only values
for system address bits A[26:19] are used giving a possible starting address from 0−128 Mbytes on a 0.5-Mbyte
boundary. The SMBA end address is implied from the top-of-memory value, or just before the beginning of an
enabled PCI Hole (bits [7:6] of DRAMC Register) at 14 or 15 Mbytes, when top-of-memory is at 16 Mbytes.
NOTE
The SMBSA Register should never be programmed with the values 00h or 01h (lower 1-Mbyte area). The
default value of 0Eh is for a 1-Mbyte SMBA area in an 8-Mbyte system DRAM configuration.
Example A: For a shared memory buffer size of 2.5 Mbytes in a 16 -Mbyte system with no PCI Hole
enabled, the SMBA value is set to a 13.5-Mbyte start address value or 1Bh.
Example B: For a shared memory buffer size of 1 Mbyte in a 16 -Mbyte system with a 1-Mbyte PCI
Hole enabled, the SMBA value is set to a 14-Mbyte start address value (top-of-DRAM
minus 2 Mbytes) or 1Ch.
Bit
Description
7:0
SMB Start Address (SMBSA). Bits [7:0] correspond to A[26:19], respectively.
3.2.27. GCLTGRAPHICS CONTROLLER LATENCY TIMER
Address Offset: 78h
Default Value: 23h
Access: Read/Write
The GCLT Register controls the graphics controller MREQ# (high priority GC request) to MGNT# maximum
latency. This register is used to program the Latency timer for PCI read cycles to DRAM and CPU/PCI master
single write cycles to DRAM.
Bit
Description
7:6
Reserved.
5:3
GC Latency For PCI Reads (GCLTR). This field is used to program the GC latency timer for PCI
master read cycles to DRAM. The DRAM cycle breaks to allow the GC accesses to DRAM when the
DRAM-to-PCI buffers are full or when this timer times-out, which ever occurs first.
Bits [5:3] Latency (In HCLKs) Bits [5:3] Latency (In HCLKs)
000 0 100 16 (default)
001 4 101 20
010 8 110 24
011 12 111 28
2:0
GC Latency For CPU/PCI Writes (GCLTW). These bits are used to program the GC latency timer
for PCI master single write and CPU single write cycles to DRAM. The DRAM cycle breaks to allow
the GC access to DRAM when this timer times-out, or the DWBs become empty, or a burst write is
encountered before the timer times out, which ever occurs first.
Bits [2:0] Latency (In HCLKs) Bits [2:0] Latency (In HCLKs)
000 0 100 16
001 4 101 20
010 8 110 24
011 12
(default)
111 28
E82437VX (TVX) AND 82438VX (TDX)
45
PRELIMINARY
4.0. FUNCTIONAL DESCRIPTION
This section describes the host, PCI, secondary cache, and DRAM interfaces, The section also describes
system arbitration and the Shared Memory Buffer Architecture (SMBA).
4.1. Host Interface
The Host Interface of the TVX is designed to support the Pentium processor with bus speeds of 50 MHz, 60
MHz, and 66 MHz. The Intel 430VX PCIset supports the Pentium processor with a full 64-bit data bus, 32-bit
address bus, and associated internal write-back cache logic. Host bus addresses are decoded by the TVX for
accesses to main memory, PCI memory, and PCI I/O. The TVX also supports the pipelined addressing
capability of the Pentium processor.
4.2. PCI Interface
The TVX integrates a high performance interface to the PCI local bus taking full advantage of the high bandwidth
and low latency of PCI. The TVX is fully PCI 2.1 Compliant. Five PCI masters are supported by the integrated
arbiter including a PCI-to-ISA bridge and four general PCI masters. The TVX acts as a PCI master for CPU
accesses to PCI. The PCI Bus is clocked at one half the frequency of the CPU clock. This divided synchronous
interface minimizes latency for CPU-to-PCI cycles and PCI-to-main memory cycles.
The TVX/TDXs integrate posted write buffers for CPU memory writes to PCI. Back-to-back sequential memory
writes to PCI are converted to burst writes on PCI. This feature allows the CPU to continue posting DWord
writes at the maximum bandwidth for the Pentium processor for the highest possible transfer rates to the
graphics frame buffer.
Read prefetch and write posting buffers in the TVX/TDXs enable PCI masters to access main memory at up to
120 MB/second. The TVX incorporates a snoop-ahead feature that allows PCI masters to continue bursting on
both reads and writes even as the bursts cross cache line boundaries.
4.3. Secondary Cache Interface
The TVX integrates a high performance write-back second level cache controller using internal/external tags and
provides a full first level and second level cache coherency mechanism. The second level cache is direct
mapped, non-sectored, and supports a write-back cache policy. Cache lines are allocated on read misses (no
write allocate).
The second level cache can be configured for either 256-Kbyte or 512-Kbyte cache sizes using Pipelined
Synchronous Burst SRAMs, DRAM Cache, or Asynchronous SRAM. The DRAM Cache is a replacement for the
standard pipeline burst SRAM. One additional PCIset pin (KRQAK), two system signals (H/WR# and RESET#),
and one DRAM Cache specific signal (M/S#; strapped at the DRAM Cache device) are required to support
DRAM Cache. The DRAM Cache conforms to the standard pipeline burst footprint. Refer to the Pipeline burst
and DRAM Cache diagrams below for additional information.
For the 256-Kbyte configurations, an 8kx8 standard Asynchronous SRAM is used to store the tags. For the 512-
Kbyte configurations, a 16kx8 standard Asynchronous SRAM is used to store the tags and the valid bits.
A second level cache line is 32 bytes wide. In the 256-Kbyte configurations, the second level cache contains 8k
lines, while the 512-Kbyte configurations contain 16k lines. Valid and modified status bits are kept on a per-line
basis. Cacheability of the entire memory space in first level cache is supported For the second level cache, only
the lower 64 Mbytes of main memory are cacheable (only main memory controlled by the TVX DRAM interface
is cached). PCI memory is not cached. The DRAM region between the SMB start address register and Top of
Memory is also not cached when SMB (Shared memory buffer) is enabled. Table 8 shows the different standard
SRAM access time requirements for different host clock frequencies.
82437VX (TVX) AND 82438VX (TDX) E
46
PRELIMINARY
Table
8
. SRAM Access Time Requirements
Host Clock Frequency
(MHz)
Std SRAM Access
Time (ns)
Burst SRAM Clock-to-
Output Access Time (ns)
Tag RAM Cycle Time (ns)
50 20 20 20
60 17 15 15
66 15 9 15
Figure 3 through Figure 5 show the connections between the TVX and the external tag RAM and data SRAM.
TIO[7:0]
TWE#
GWE#
CA[4:3]
D[7:0]
WE#
OE#A[13:0]
8Kx8 Tag RAMHA[17:5]
HD[63:56]
HD[47:40]
HD[39:32]
HD[31:24]
HD[23:16]
HD[15:8]
HD[7:0]
TVX
32Kx8 SRAM
COE#
BWE#/
CGCS#
HD[55:48]
HBE[7:0]#
(from
CPU)
F08 x 2
D[7:0]
CS#
D[7:0]
CS#
D[7:0]
CS#
D[7:0]
CS#
D[7:0]
CS#
D[7:0]
CS#
D[7:0]
CS#
A[1:0]
WE#
D[7:0]
OE#
CS#
A[14:2]
055305
Figure 3. TVX Connections to 256-Kbyte Second Level Cache with Asynchronous SRAM
E82437VX (TVX) AND 82438VX (TDX)
47
PRELIMINARY
D[7:0]
D[15:8]
D[23:16]
D[31:24]
TIO[7:0]
TWE#
CCS#
COE#
GWE#
BWE#
KRQAK
D[7:0]
WE#
OE#
A[12:0]
8Kx8 Tag RAMHA[17:5]
HD[63:56]
HD[55:48]
HD[47:40]
HD[39:32]
HD[31:24]
HD[23:16]
HD[15:8]
HD[7:0]
TVX
32Kx32 SRAM
HCLK
HA[17:3]
CADS#
CADV#ADS#
CLK
A[14:0]
CS#
GWE#
D[7:0]
OE#
ADSC#
ADV#
ADSP#
BWE#
D[15:8]
KRQAK
BWE[3:0]#
D[23:16]
D[31:24]
M/S#
HBE[7:4]#
BWE[3:0]#
KRQAK
GWE#
BWE#
HBE[3:0]#
HW/R#
RESET#
RESET#
M/S#
HW/R#
VCC
HW/R#
HW/R#
CPURST#
055306
NOTES:
KRQAK (and the 4.7KΩ resistor on the KRQAK signal), M/S#, RESET# (and inverter on the RESET# signal), and H/WR# are
only present for DRAM Cache Mode. M/S# is a master/slave strapping requirement used by the DRAM Cache to determine
which device will drive the KRQAK signal during a refresh. The device that is strapped high (master), is the device that drives
KRQAK. All other DRAM Cache devices monitor this signal.
Figure 4. TVX Connections to 256-Kbyte Second Level Cache with Pipelined Burst
82437VX (TVX) AND 82438VX (TDX) E
48
PRELIMINARY
TIO[7:0]
TWE#
CCS#
COE#
GWE#
BWE#
KRQAK
HA[18:5]
TVX
HA[17:3]
CADS#
CADV#
ADS#
HA18
GND
HD[63:56]
HD[55:48]
HD[47:40]
HD[39:32]
HD[31:24]
HD[23:16]
HD[15:8]
HD[7:0]
HA18
D[7:0]
WE#
OE#A[12:0]
16Kx8 Tag RAM
D[7:0]
D[15:8]
D[23:16]
D[31:24]
HD[63:56]
HD[55:48]
HD[47:40]
HD[39:32]
HD[31:24]
HD[23:16]
HD[15:8]
HD[7:0]
GWE#
D[7:0]
BWE#D[15:8]
HBE[3:0]#
KRQAK
D[23:16]
D[31:24]
HW/R#
RESET#
M/S#
HW/R#
VCC
CPURST#
HCLK
32Kx32 SRAM
CLK
A[14:0]
CS1#
GWE#
D[7:0]
OE#
ADSC#
ADV#
ADSP#
BWE#
D[15:8]
HBE[3:0]#
KRQAK
D[23:16]
D[31:24]
CS2
CS2#
32Kx32 SRAM
HCLK
CCS#
COE#
CADS#
CADV#
HA[17:3]
ADS#
VDD
CLK
A[14:0]
CS1#
D[7:0]
OE#
ADSC#
ADV#
ADSP#
D[15:8]
D[23:16]
D[31:24]
CS2
CS2#
HBE[7:4]#
HBE[3:0]#
GWE#
BWE#
HBE[3:0]#
KRQAK
GWE#
BWE#
HBE[3:0]#
KRQAK
HBE[3:0]#
HBE[7:4]#
GWE#
BWE#
M/S#
HW/R#
RESET#
M/S#
HW/R#
RESET#
HW/R#
M/S#
HW/R#
RESET#
HW/R#
HW/R#
055307
NOTES:
KRQAK (and the 4.7KΩ resistor on the KRQAK signal), M/S#, RESET# (and inverter on the RESET# signal), and H/WR# are
only present for DRAM Cache Mode. M/S# is a master/slave strapping requirement used by the DRAM Cache to determine
which device will drive the KRQAK signal during a refresh. The device that is strapped high (master), is the device that drives
KRQAK. All other DRAM Cache devices monitor this signal.
Figure 5. TVX Connections for 512-Kbyte Second Level Cache with Pipelined Burst SRAM
E82437VX (TVX) AND 82438VX (TDX)
49
PRELIMINARY
4.3.1. CLOCK LATENCIES
Table 9 and Table 10 list the latencies for various processor transfers to and from the second level cache.
Table
9
. Second Level Cache Latencies (Asynchronous SRAM)
Cycle Type
HCLK Count
Burst Read
3-2-2-2
Burst Write (write-back)
4-3-3-3
Single Read
3
Single Write
4
Back-to-Back Burst Reads
3-2-2-2,3-2-2-2 (note 1)
NOTE:
1. The back to back cycles do not account for CPU idle clocks between cycles.
Table
10
. Second Level Cache Latencies (Pipelined Burst SRAM or DRAM Cache)
Cycle Type
HCLK Count
Burst Read
3-1-1-1
Burst Write (write-back)
3-1-1-1
Single Read
3
Single Write
3
Pipelined Back-to-Back Burst Reads
3-1-1-1,1-1-1-1 (note 1)
NOTE:
1. The back to back cycles do not account for CPU idle clocks between cycles.
4.3.2. SNOOP CYCLES
Snoop cycles are used to maintain coherency between the caches (first and second level) and main memory.
The TVX generates a snoop (or inquire) cycle to probe the first level and second level caches when a PCI
master attempts to access main memory. Snoop cycles are performed by driving the PCI master address onto
the host address bus and asserting EADS#.
To maintain optimum PCI bandwidth to main memory, the TVX utilizes a "snoop ahead" algorithm. Once the
snoop for the first cache line of a transfer has completed, the TVX automatically snoops the next sequential
cache line. This algorithm enables the TVX to continue burst transfers across cache line boundaries.
Reads
If the snoop cycle generates a first level cache hit to a modified line, the line in the first level cache is written
back to main memory (via the DRAM posted write buffers). The line in the second level cache (if it exists) is
invalidated. Note that the line in the first level cache is not invalidated if the INV pin on the CPU is tied to the
KEN# signal from the TVX. The TVX drives KEN#/INV low with EADS# assertion during PCI master read cycles.
At the same time as the first level snoop cycle, the TVX performs a tag look-up to determine whether the
addressed memory is in the second level cache. If the snoop cycle generates a second level cache hit to a
modified line and there was not a hit in the first level cache (HITM# not asserted), the second level cache line is
82437VX (TVX) AND 82438VX (TDX) E
50
PRELIMINARY
written back to main memory (via the DRAM posted write buffers) and changed to the "clean" state. The PCI
master read completes after the data has been written back to main memory.
Writes
PCI Master write cycles never result in a write directly into the second level cache. A snoop hit to a modified line
in either the first or second caches results in a write-back of that line to main memory. If both the first and second
level caches have modified lines, the line is written back from the first level cache. In all cases, lines in the first
and second level caches are invalidated and the PCI write to main memory occurs after the writeback
completes. A PCI master write snoop hit to an unmodified line in either the first or second level caches results in
the line being invalidated. The TVX drives KEN#/INV with EADS# assertion during PCI master write cycles.
4.4. DRAM Interface
The TVX integrates a DRAM controller that supports a 64-bit memory array from 4 to 128 Mbytes of main
memory. Extended Data Out (EDO), standard page mode (SPM), and synchronous DRAM (SDRAM) are
supported. The TVX does not support parity and requires that non-parity SIMMS and DIMMS be used. The TVX
generates the DRAM control signals and multiplexed addresses for the DRAM array and controls the data flow
through the TDXs. For CPU-to-DRAM cycles the address flows through the TVX and data flows through the
TDXs. For PCI or ISA cycles to memory, the address flows through the TVX and data flows to the TDXs through
the TVX and PLINK bus. The TVX and TDX DRAM interfaces are synchronous to the CPU clock. The DRAM
controller interface is fully configurable through a set of control registers. Complete descriptions of these
registers are given in the TVX configuration register description. A brief overview of the registers that configure
the DRAM interface is provided in this section.
The TVX supports industry standard 32-bit wide memory modules with EDO and SPM. The twelve multiplexed
address lines (MA[11:0]) allow the TVX to support 512Kx32, 1Mx32, 2Mx32, and 4Mx32 SIMM's, with both
symmetrical and asymmetrical addressing. Five RAS# lines support up to five rows of DRAM. If the 5th RAS#
line is supported, certain restrictions are placed on the system (refer to the DRAM Organization section), Eight
CAS# lines allow byte control over the array. The TVX targets 60 ns DRAMs and supports both single and
double-sided SIMM's. The TVX also provides an automatic CBR refresh, at a rate of 1 refresh per 15.6
microseconds at 66, 60, and 50 MHz.
The TVX supports unbuffered DIMM Modules with SDRAMs. The twelve multiplexed address lines (MA[11:0])
allow the TVX to support 1Mx64, 2Mx64, and 4Mx64 DIMM's. Five CS# lines support up to five rows of SDRAM.
If the 5th CS# line is supported, certain restrictions are placed on the system (refer to the DRAM Organization
section). Eight DQM lines allow byte control over the array. The TVX supports 50, 60 and 66 MHz SDRAMs and
supports both single and double-sided DIMM's. The TVX also provides an automatic CBR refresh, at a rate of 1
refresh per 15.6 microseconds at 66, 60, and 50 MHz.
The DRAM interface is configured by the DRAM Control (DRAMC) Register, the DRAM Extended Control
(DRAMEC) Register, DRAM Timing (DRAMT) Register, SDRAM Control (SDRAMC) Register, the four DRAM
Row Boundary (DRB) Registers, and the DRAM Row Type (DRT) Registers. The four DRB Registers define the
size of each row in the memory array, enabling the TVX to assert the proper RAS# line for accesses to the
array.
Seven Programmable Attribute Map (PAM) Registers specify the cacheability, PCI enable, and read/write status
of the memory space between 640 Kbytes and 1 Mbyte. Each PAM Register defines a specific address area
enabling the system to selectively mark specific memory ranges as cacheable, read-only, write-only, read/write,
or disabled. When a memory range is disabled, all CPU accesses to that range are forwarded to PCI.
The TVX also supports one of two memory holes (512-Kbyte to 640-Kbyte or 14/15 Mbytes to 16 Mbytes).
Accesses to the memory holes are forwarded to PCI. The memory hole can be enabled/disabled through the
DRAM Control Register. All other memory from 1 Mbyte to the top of memory is read/write and cacheable.
The SMRAM memory space is controlled by the SMRAM Control Register. This register enables, opens, closes,
or locks SMRAM space.
E82437VX (TVX) AND 82438VX (TDX)
51
PRELIMINARY
4.4.1. DRAM ORGANIZATION
The Intel 430VX PCIset supports EDO, SPM, and SDRAM. In a four row system, all of these DRAM types can
be mixed and matched on a row-by-row basis. In a 5 row system, EDO/SPM can not be mixed with SDRAM.
Refer to the “Four Row System” and “Five Row System” sections below, for specific details.
NOTE
SDRAM is not supported in the row the graphics controller uses in a Shared Memory Buffer Architecture
(SMBA) configuration. SDRAM with EDO is supported in a SMBA configuration, if SDRAM is not
populated in the SMBA DRAM row.
Four Row System
Figure 6 illustrates a 4-SIMM configuration that supports 4 double-sided SIMM's. A row in the DRAM array is
made up of two SIMM's that share a common RAS# line. Within any given row, the two SIMMs must be of the
same size. Among the four rows, SIMM densities can be mixed in any order. Each row is controlled by up to 8
CAS lines. EDO and standard page mode DRAM's can be mixed between rows; however, a given row must
contain only one type of DRAM. When DRAM types are mixed (EDO and standard page mode) each row runs
optimized for that particular type of DRAM. Figure 8 illustrates a 2-DIMM SDRAM system.
The rules for SIMM population are as follows:
•SIMM sockets can be populated in any order (i.e., memory for RAS0# does not have to be populated before
memory for RAS3# is used).
•SIMM socket pairs need to be populated with the same densities on a row per row basis. For example,
SIMM sockets for RAS0# should be populated with identical densities. However, SIMM sockets for RAS3#
can be populated with different densities than the SIMM socket pair for RAS0#.
•EDO's and standard page mode can both be used on a row per row basis. However, only one type should
be used per SIMM socket pair. For example, SIMM sockets for RAS[0]# can be populated with EDO's while
SIMM sockets for RAS[3]# can be populated with standard page mode.
The rules for SDRAM population are as follows:
•DIMMs can be populated in any order ( e.g., row 0 does not need to be populated before row 3 is used).
•SDRAMs can be mixed with EDO/SPM on a row per row basis (e.g., row 0 can be populated with SDRAMs
while row 3 can be populated with EDO/SPM). TVX runs with optimized timings on a row per row basis.
82437VX (TVX) AND 82438VX (TDX) E
52
PRELIMINARY
32-Bit SIMM 32-Bit SIMM
WEB#
WEA#
MA[11:0]
32-Bit SIMM 32-Bit SIMM
CAS[7:4]#
CAS[3:0]#
TDXTDX
MD[63:56,47:40,
31:24,15:8] MD[55:48,39:32,
23:16,7:0]
RAS[3:0]#
RAS[3:2]#
RAS[1:0]#
PLINK[15:0]
PLINK[15:8]
PLINK[7:0]
HD[63:56,47:40,
31:24,15:8] HD[55:48,39:32,
23:16,7:0]
Host Data Bus
TVX
055308
NOTES:
1. External buffers on the MA lines have been integrated in the Intel 430VX PCIset.
2. Two copies of WE# are provided for loading purposes only.
Figure 6. SIMM Socket With EDO/SPM
E82437VX (TVX) AND 82438VX (TDX)
53
PRELIMINARY
64-Bit DIMM
WEB#
WEA#
MA[11:0]
64-Bit DIMM
DQM[7:4]#DQM[3:0]#
TDXTDX
MD[63:56,47:40,
31:24,15:8] MD[55:48,39:32,
23:16,7:0]
CS[3:0]#
CS[3:2]#
CS[1:0]#
PLINK[15:0]
PLINK[15:8]
PLINK[7:0]
HD[63:56,47:40,
31:24,15:8] HD[55:48,39:32, 23:16,7:0]
Host Data Bus
TVX
SRASB#
SCASB#
SRASA#
SCASA#
055309
NOTES:
1. External buffers on the MA lines have been integrated in the Intel 430VX PCIset.
2. Two copies of WE#, SRAS#, and SCAS# are provided for loading purposes only.
Figure 7. DIMM Socket With SDRAM
82437VX (TVX) AND 82438VX (TDX) E
54
PRELIMINARY
Five ROW System
Support for a 5th row enables memory to be soldered on the motherboard, in addition to providing 4 rows of
memory upgradability. Mixing of memory types is allowed on a row per row basis for a system with 4 rows of
memory. If the 5th row is used, certain restrictions are placed on the system. These restrictions are shown
below:
•The 5th row is intended to be implemented with DRAM devices that are soldered down on the motherboard.
If a DIMM or a SIMM is used in the 5th ROW, it should not be used as an upgrade path by the end user, as
the size and type of DRAM that can be implemented in the 5th row is limited. (see the following bullets).
•All 5 rows must be either all EDO/SPM type DRAM or all SDRAM (i.e., EDO and SPM can be mixed on a
row per row basis but EDO/SPM cannot be mixed with SDRAMs).
•There is a restriction on DRAM timings when 5 rows of memory are populated. At 66 MHz with 60 ns
memory or 60 MHz with 70 ns memory, the burst rate must be set to X-3-3-3 EDO, x-4-4-4 FPM
(bits[6:5]=01, DRAMT register, 58h).
•In a SMBA design, the size of the 5th row of memory must be fixed at 8 Mbytes using 1Mx16 devices, due to
loading limitations. In a non-SMBA design, the 5th row can be implemented using 2Mx8 (fixed at 16M) or
1Mx16 devices (fixed at 8M). Note that the 16M in the 5th row is supported in non-SMBA designs only.
•Note that the total memory supported is limited to 128 Mbytes, even though it is possible to populate the
5 rows with up to 136 Mbytes (in a SMBA design) or 144 Mbytes (in a non-SMBA design). This limit
must be ensured by the system BIOS.
If the first four rows have 32 Mbytes each and the 5th row has 8 Mbytes of memory for a total of 136
Mbytes, BIOS must program one of the first four rows with 24 Mbytes to ensure that the 5th row is fully
accessable.
If the first four rows of memory have 32 Mbytes each, and the 5th row has 16 Mbytes, BIOS must
disable the 5th row of memory via the DRB’s. Also, if the 5th row of memory has 16 Mbytes and the
other four rows of memory have 1Mx4 (8 Mbytes per row) or 4Mx4 (32 Mbytes per row), BIOS must add
wait-states to the memory array. The wait-states added slows the read and write leadoff and burst
timings by one clock.
For a 5 row SPM/EDO DRAM implementation, SCASB#/RAS4# is used as the RAS4# signal to provide the
5th RAS# signal. For a 5 row SDRAM implementation, MXS/CS4# is used as the CS4# signal to provide the
5th CS# signal. Both RAS4# and CS4# have an internal pullup resistor in the TVX.
Figure 8 shows an EDO/FPM system with 5 rows of memory, using 1Mx16 devices (8M). If 2Mx8 devices are
used (fixed at 16M), the WEB# line can be connected to four devices in the 5th row, and the WEB# can be
connected to the remaining four devices in the 5th row. In general, the WEA# and WEB# lines should be
distributed as equally as possible in the DRAM subsystem.
E82437VX (TVX) AND 82438VX (TDX)
55
PRELIMINARY
1Mx16
32-Bit SIMM 32-Bit SIMM
WEB#
WEA#
MA[11:0]
CAS[7:4]#CAS[3:0]#
TDXTDX
MD[63:56,47:40,
31:24,15:8]
MD[55:48,39:32, 23:16,7:0]
RAS[3:0]#
RAS[3:2]#
RAS[1:0]#
PLINK[15:0]
PLINK[15:8]
PLINK[7:0]
HD[63:56,47:40,
31:24,15:8]HD[55:48,39:32, 23:16,7:0]
Host Data Bus
TVX
32-Bit SIMM 32-Bit SIMM
1Mx16 1Mx16 1Mx16
RAS4#
WEB#
055310
Figure 8. 5 ROW Example using EDO/FPM
Table 11 provides a summary of the characteristics of memory configurations supported by the TVX. Minimum
values listed are obtained with single-sided SIMMs and maximum values are obtained with double-sided SIMMs.
Note that, for a 64-bit wide memory array, a minimum of two 32-bit wide DRAM SIMM configuration is required.
The minimum values used are also the smallest upgradable memory size.
82437VX (TVX) AND 82438VX (TDX) E
56
PRELIMINARY
Table
11
. Minimum (Upgradable) and Maximum Memory Size for each configuration (EDO/SPM)
DRAM DRAM DRAM
DRAM SIMM
DRAM
Address Size
DRAM Size
Tech.
Density
Width
SS
x32
DS
x32
Addressing
Row
Column
Min.
(1 row)
Max.
(4 rows)
4Mbit 512K 8/32 512K 1M
Asymmetric
10 9
4 MB
16 MB
1M 41M 2M
Symmetric
10 10
8 MB
32 MB
1M 41M 2M
Asymmetric
12 8
8 MB
32 MB
16Mbit 1M 16 1M 2M
Symmetric
10 10
8 MB
32 MB
1M 16 1M 2M
Asymmetric
12 8
8 MB
32 MB
2M 82M 4M
Asymmetric
11 10
16 MB
64 MB
4M 44M 8M
Symmetric
11 11
32 MB
128 MB
4M 44M 8M
Asymmetric
12 10
32 MB
128 MB
Table 12 provides a summary of the characteristics of memory configurations supported by the TVX. Minimum
values listed are obtained with single-sided DIMMs and maximum values are obtained with double-sided DIMMs.
The minimum values used are also the smallest upgradable memory size.
Table
12
. Minimum (Upgradable) and Maximum Memory Size for each configuration(SDRAM)
SDRAM SDRAM SDRAM
DRAM DIMM
DRAM
Address Size
DRAM Size
Tech.
Density
Width
SS
x64
DS
x64
Addressing
Row
Column
Min.
(1 row)
Max.
(4 rows)
16Mbit 1M 16 1M 2M
Asymmetric
12 8
8 MB
32 MB
2M 82M 4M
Asymmetric
12 9
16 MB
64 MB
4M(note)
44M 8M
Asymmetric
12 10
32 MB
128 MB
The memory organization (Figure 9) represents the maximum 128 Mbytes of address space. Accesses to
memory space above top of DRAM, video buffer, or the memory gaps (if enabled) are forwarded to PCI. These
regions are not cacheable. Below 1 Mbyte there are several memory segments that have selectable
cacheability. The DRAM space occupied by the video buffer or the memory space gaps is not remapped and is
therefore "lost". Refer to the Shared Memory Buffer Architecture (SMBA) section for details on SMBA System
memory mapping.
E82437VX (TVX) AND 82438VX (TDX)
57
PRELIMINARY
4 GB
128 MB
64 MB
16 MB
1 MB
768 KB
640 KB
0
15 MB
Always Cacheable
512 KB
Optional Memory Space Gap
Optional Memory Space Gap
Cacheable
DOS Applications
(No Read/Write Protect)
(Always Cacheable)
Forward To PCI
(Non-Cacheable)
Expansion and BIOS Region
(Cacheable Segments)
Video Buffer
(SMM Space Non-Cacheable)
Second Level Cache (Non-Cacheable)
First Level Cache (Cacheable)
14 MB
055311
Figure 9. Memory Space Organization
4.4.2. DRAM ADDRESS TRANSLATION
The multiplexed row/column address to the DRAM memory array is provided by the MA[11:0] signals. The
MA[11:0] bits are derived from the host address bus as defined by Table 13. The TVX has a 2-Kbyte page size.
However, the address map has been defined to support a 4k page size from the graphics controller’s
perspective, in a SMBA design. The page offset address is driven over the MA[7:0] lines when driving the
column address. The MA[11:0] lines are translated from the host address lines A[24:3] for all memory accesses.
Table
13
. DRAM Address Translation for EDO/SPM/SDRAM
MA[11:0]
11 10 9 8 7 6 5 4 3 2 1 0
Row A11 A22 A21 A20 A19 A18 A17 A16 A15 A14 A13 A12
Column A11 A11/
”V”
A11/
A24
A11/
A22/
A23
A10 A9 A8 A7 A6 A5 A4 A3
NOTES:
V=Valid level (either 0 or 1) used for SDRAMs. During the initialization sequence, it is 1 and during normal mode of operation, it
is 0.
82437VX (TVX) AND 82438VX (TDX) E
58
PRELIMINARY
4.4.3. DRAM PAGING
The DRAM page is kept open when the CPU host bus is non-idle or the PCI interface owns the bus.
4.4.4. EDO MODE
Extended Data Out (or Hyper Page Mode) DRAM is designed to improve the DRAM read performance. EDO
DRAM holds the memory data valid until the next CAS# falling edge. Note that standard page mode DRAM tri-
states the memory data when CAS# negates to precharge. With EDO, the CAS# precharge overlaps the
memory data valid time. This allows CAS# to negate earlier while still satisfying the memory data valid window
time.
4.4.5. SDRAM MODE
The Intel 430VX PCIset supports all of the features and timings as shown in the “SDRAM PC” specification. The
objective of the “SDRAM PC” specification is to enable low cost and easily manufacturable SDRAMs for the
main stream volume desktop PC’s. There are three speed grade parts defined for the Intel 430VX PCIset
(Table 14). All of the speed grades conform to the “SDRAM PC” specification. The three speed grades are as
follows:
Table
14
. SDRAM Speed Grades
Speed Grade
CAS latency
System Frequency
66.67 MHz (note)
3
50/60/66 MHz
66.67 MHz
2
50/60/66 MHz
50 MHz
2
50 MHz
SDRAM Mode Register
The Mode Register Set supported by TVX is as follows:
Bit 11 10 9 8 7
6, 5, 4
3
2, 1, 0
Value
0 0 0 0 0 CL WT BL
CAS Latency Field (CL)
Wrap Type (WT)
Burst Length (BL)
Bits (6:4)
CAS Latency
Bit 3
Type
Bits (2:0)
Burst
Length
010 2 0 X010 4
011 3 1
Interleave
All Other
X
All Other
X
NOTES:
1. The CL bit is programmed as per the bit setting for the CL field of the TVX SDRAMC Register
2. The linear order addressing is not supported.
3. X=Don’t Care. These modes are Don’t Care for TVX specific implementation.
E82437VX (TVX) AND 82438VX (TDX)
59
PRELIMINARY
Command Truth Table
SDRAM commands supported by the TVX are:
•Mode Register Set (MRS)
•Activate Bank (ACT)
•Read Bank (READ)
•Write Bank (WRITE)
•Precharge All Banks (PALL)
•Deselect Device
•No Operation (NOP)
•Auto Refresh (REFR)
4.4.6. AUTO DETECTION
The SDRAM, FPM, and EDO detection is performed by BIOS. Note, that when accessing any of the DRAM
related registers (i.e., 54h−68h), refresh should be turned off via the DRAMC Register.
4.4.7. DRAM PERFORMANCE
The DRAM performance is controlled by the DRAM timing register, processor pipelining, and by the type of
DRAM used (EDO or standard page mode). Table 15 depicts both EDO and standard page mode optimum
timings. For read cycles, clocks counts are measured from ADS# to BRDY#. For write cycles, the measurement
is broken up into two parts. The first part consists of the rate of posting data in to the CPU to DRAM posted write
buffers. This is measured from ADS# to BRDY#. The second part consists of the retire rate from posted write
buffers to the DRAM. The leadoff for retiring is measured from the clock after BRDY# assertion to the CAS#
assertion. Table 15 lists a performance summary for 60 ns EDO/SPM DRAMs.
Table
15
. EDO/ Standard Page Mode Performance Summary (60ns DRAMs)
Processor Cycle
Type (pipelined)
66 MHz
7
(Burst SRAM)
60 MHz
(Burst SRAM)
50 MHz
(Burst SRAM)
66 MHz
(Asynch
Cache)
DRAM
TYPE
Burst Read Page Hit
6-2-2-2
5-2-2-2
5-2-2-2
7-3-3-3
EDO
Read Row Miss
1
9-2-2-2
8-2-2-2
7-2-2-2
10-3-3-3
EDO
Read Page Miss
12-2-2-2
11-2-2-2
10-2-2-2
13-3-3-3
EDO
Back-to-Back Burst
Reads Page Hit
6-2-2-2-3-2-2-2
5-2-2-2-3-2-2-2
5-2-2-2-3-2-2-2
7-3-3-3-7-3-3-3
EDO
Burst Read Page Hit
6-3-3-3
6-3-3-3
6-3-3-3
7-3-3-3
SPM
Burst Read Row Miss
1
9-3-3-3
9-3-3-3
8-3-3-3
10-3-3-3
SPM
Burst Read Page Miss
12-3-3-3
12-3-3-3
11-3-3-3
13-3-3-3
SPM
Back-to-Back Burst
Read Page Hit
6-3-3-3-3-3-3-3
6-3-3-3-3-3-3-3
6-3-3-3-3-3-3-3
7-3-3-3-7-3
SPM
Write Page Hit2,3,4
3 3 2 3 EDO/SPM
Write Row Miss2,3,4
6 6 5 6 EDO/SPM
Write Page Miss2,3,4
9 9 8 9 EDO/SPM
Posted Write3,4
3-1-1-1
3-1-1-1
3-1-1-1
4
-1-1-1
EDO/SPM
82437VX (TVX) AND 82438VX (TDX) E
60
PRELIMINARY
Table
15
. EDO/ Standard Page Mode Performance Summary (60ns DRAMs)
Processor Cycle
Type (pipelined)
66 MHz
7
(Burst SRAM)
60 MHz
(Burst SRAM)
50 MHz
(Burst SRAM)
66 MHz
(Asynch
Cache)
DRAM
TYPE
Write retire rate from
Posted Write Buffer
-2-2-2
-2-2-2
-2-2-2
- -2-2-2
EDO/SPM
Single writes
2 2 2 2 EDO
Single writes
2 2 2 2 SPM
DRAMEC Register (offset 56h) Programming
Bit 6 (RRA)
0 0 0 0 EDO/SPM
Bit 5(FEPS)
0 1 1 0 EDO
Bit 5(FEPS)
0 0 0 0 SPM
DRAMT Register (offset 58h) Programming
Bit 7(FMRD)
1 1 1 1 EDO/SPM
Bits[6:5] (DRBT)
62 2 2 2 EDO/SPM
Bit[4:3] (DWBT)
2 2 2 2 EDO/SPM
Bit 2 (FRCD)
0 0 1 0 EDO/SPM
Bits(1:0) (DLT)
1 1 1 1 EDO/SPM
NOTES:
1. The above row miss cycles assume that the new page is closed from the prior cycle. Due to the MA[11:0] to RAS# setup
requirements, if the page is open, 2 clocks are added to the leadoff. If FMRD bit is set to 1, then only one clock is added to
the leadoff.
2. This cycle timing assumes the write buffer(DWB) is empty.
3. Write timing is measured from the clock after BRDY# is returned to the CPU up to CAS# assertion for that cycle.
4. Write data is always posted as 3-1-1-1 (ADS# to BRDY#, if buffer is empty) with burst or pipelined burst second level
cache SRAMs (or cacheless operation). For Asynchronous SRAMs, write data is posted as 4-1-1-1.
5. Bit 5 in the DRAMEC register must be set to 0 if the EDO/FPM burst rate is set to x333 or x444, or an asynchronous
cache in the system.
6. If an L2 asynchronous cache is in the system, do not set the DRBT bit to 3.
7. There is a restriction on the DRAM timings when 5 rows of memory are populated. At 66 MHz with 60 ns memory or 60
MHz with 70 ns memory, the burst rate must be set to x-3-3-3 EDO, x-4-4-4 SPM (register 58h, bits[6:5]=01h.
E82437VX (TVX) AND 82438VX (TDX)
61
PRELIMINARY
Table
16
. EDO AND SPM LEADOFF TIMING CALCULATION
DLT1
Read
leadoff
Write
Leadoff
RAS#
Precharge
FRCD2
Fast
RAS to
CAS
Leadoff
FEPS3
Leadoff
0 11 7 3 0 3 0 0 0
1 10 6 3 1 2 -1 1-1
2 11 7 4
3 10 6 4
NOTES:
1. DLT bits represent the row miss leadoff for EDO/SPM reads and writes. These leadoff represent timings with FRCD=0,
SLE=0 and FEPS=0. Page hits can be calculated by subtracting RAS# precharge from row misses (10 - 3 =7) and page
misses can be calculated by adding the RAS# precharge time to Row Misses.
DLT effects the row miss and page miss timings only (e.g., DLT= 1 is one clock faster than DLT=0 on row miss and page
miss timings). These bits control MA setup to CAS# assertion.
DLT does not effect page hit timings (e.g., DLT=0 or DLT=1 have same page hit timings for reads and writes). In both
cases it would be 7 (10-3) clocks.
2. FRCD bit effects the row miss and page miss timings for EDO/SPM reads and writes. FRCD bits have subtractive effect
on DLT bits (e.g., if DLT=1 and FRCD=1, read row miss leadoff timing is 9 (10 -1) clocks).
3. FEPS bit effects the page miss, row miss and page hit timings for EDO reads only. FEPS bits have subtractive effect on
DLT bits (e.g., if DLT=1 and FEPS=1, then read row miss leadoff timing is 9 (10-1) clocks).
Table
17
. SDRAM Performance Summary
Processor Cycle type
50/60/66 MHz
50/60/66 MHz
50 MHz
CL=3
1
CL=2
2
CL=2
3
Burst Read Page Hit
7-1-1-1 6-1-1-1 6-1-1-1
Read row Miss
10-1-1-1 8-1-1-1 8-1-1-1
Read Page Miss
13-1-1-1 11-1-1-1 11-1-1-1
Back-to-Back Burst
Reads Page Hit
7-1-1-1 2-1-1-1.
6-1-1-1 2-1-1-1.
6-1-1-1 2-1-1-1.
Write Page Hit
3 3 3
Write Row Miss
6 5 5
Write Page Miss
9 8 8
Posted Write
3-1-1-1 3-1-1-1 3-1-1-1
Write retire rate from
Posted Write Buffer
-1-1-1 -1-1-1 -1-1-1
SDRAMC Register (offset 54
−−
55h) Programming
Bit 4 (CL)
0 1 1
Bit 3 (RT)
0 0 1
82437VX (TVX) AND 82438VX (TDX) E
62
PRELIMINARY
NOTES:
1. Parts are of Speed Grade 66.67 MHz with CAS latency =3
2. Parts are of Speed Grade 66.67 MHz with CAS latency =2
3. Parts are of Speed Grade 50 MHz with CAS latency =2
4. Asynch SRAMs are not supported with SDRAMs.
5. The row miss cycles assume that the new page is closed from the prior cycle.
4.4.8. DRAM REFRESH
TVX supports CBR refresh only and generates refresh requests. The rate at which requests are generated is
determined by the value in the DRAM Refresh Rate field of the DRAM Control Register. When a refresh request
is generated, it is placed in a four entry queue. DRAM high level priority refresh is set after four refreshes are
accumulated. If one of the refresh is serviced after the high level priority refresh is set, the priority level of the
Refresh changes from high to low.
There is also a "smart refresh" algorithm implemented in the refresh controller. Refresh is only performed on
banks that are populated. The refresh controller determines which banks are populated through the DRB
registers.
4.4.9. SYSTEM MANAGEMENT RAM
The TVX supports the use of main memory as System Management RAM (SMRAM) enabling the use of System
Management Mode. When this function is disabled, the TVX memory map is defined by the DRB and PAM
Registers. When SMRAM is enabled, the TVX reserves the video buffer area (A and B segments) of main
memory for use as SMRAM.
SMRAM is placed at A0000BFFFFh via the SMRAM Space Register. Enhanced SMRAM features can also be
enabled via this register. PCI masters can not access SMRAM when it is programmed to the video buffer area (A
and B segments).
When the TVX detects a CPU stop grant special cycle, it generates a PCI Stop Grant Special cycle, with 0002h
in the message field (AD[15:0]) and 0012h in the message dependent data field (AD[31:16]) during the first data
phase (IRDY# asserted).
4.5. TDXs and the PLINK Interface
The TDXs provide the data path for host-to-DRAM, PCI-to-DRAM, and host-to-PCI cycles. Two TDXs are
required for the Intel 430VX PCIset. The TDXs are divided into two identical units. The TVX controls the data
flow through the TDXs with signals PCMD[1:0], HOE#, POE#, MOE#, MSTB[1:0], MXS, and MADV#.
The TDXs have three data path interfaces; the host bus HD[63:0]. the memory bus MD[63:0], and the
PLINK[15:0] bus between the TDX and TVX. The data paths for the TDXs are interleaved on byte boundaries
(Figure 10). Byte lanes 0, 2, 4, and 6 from the host CPU data bus connects to the even order TDX and byte
lanes 1, 3, 5, and 7 connects to the odd order TDX. PLINK[7:0] connects to the even order TDX and PLINK[15:8]
connects to the odd order TDX.
E82437VX (TVX) AND 82438VX (TDX)
63
PRELIMINARY
MD[31:0]
HD[31:0]
HOE#
MOE#
POE#
PCMD[1:0]
MSTB[1:0]#
MADV#
MXS
PLINK[7:0]
PLINK[15:8]
MD[63:56, 47:40,
31:23, 15:8]
HD[63:56, 47:40, 31:23, 15:8]
HOE#
MOE#
POE#
PCMD[1:0]
MSTB[1:0]
MADV#
MXS
Odd
Order
TXD
HCLK
MD[31:0]
HD[31:0]
HOE#
MOE#
POE#
PCMD[1:0]
MSTB[1:0]#
MADV#
MXS
PLINK[7:0]PLINK[7:0]
HD[56:48, 39:32, 23:16, 7:0]
HOE#
MOE#
POE#
PCMD[1:0]
MSTB[1:0]
MADV#
MXS
Even
Order
TXD
HCLK
MD[56:48, 39:32,
23:16, 7:0]
055312
Figure 10. TDX 64-Bit Data Path Partitioning
4.6. System Arbitration
The TVX's PCI Bus arbiter allows PCI peer-to-peer traffic concurrent with CPU main memory/second level
cache cycles. The arbiter supports five PCI masters (Figure 11). REQ[3:0]#/GNT[3:0]# are used by PCI masters
other than the PCI-to-ISA expansion bridge (PIIX3). PHLD#/PHLDA# are the arbitration request/grant signals for
the PIIX3 and provide guaranteed access time capability for ISA masters. PHLD#/PHLDA# also optimize system
performance based on PIIX3 known policies. PCI request/grant pair three is multiplexed and used as the SMBA
request/grant pair when the SMBA is enabled.
A bit in the PCON register is used to select between a priority scheme that counts three PCI grants (mode 0) or
two PCI grants (mode 1) to decide when it is time to let the host in. Mode 0 is intended to be used in a system
when SMBA is disabled and mode 1 is intended to be used in a system with SMBA.
Arbiter
PHLD#
REQ0#
REQ1#
REQ2#
REQ3#/GREQ#
PHLDA#
GNT0#
GNT1#
GNT2#
GNT3#/GGNT#
055313
Figure 11. PCI Bus Arbiter
82437VX (TVX) AND 82438VX (TDX) E
64
PRELIMINARY
4.6.1. PRIORITY SCHEME AND BUS GRANT
The TVX PCI arbitration provides two arbitration schemes (mode 0 and 1) as programmed in the PCON
Register. The highest priority requester is determined by a fixed order queue together with a highest priority
pointer. Although the priority ring is fixed, the highest priority pointer moves to determine which PCI agent is at
the top (and bottom) of the queue. The arbiter counts three grant assertions (mode 0) or two grant assertions
(mode 1) to requesters different than the one it is currently granting ( and all grants within Multi-Transaction
Timer count are collapsed to one) to determine when it’s time for host access.
The grant signals (GNTx#) are normally negated after recognition of FRAME# assertion or 16 PCLKs from grant
assertion, if no cycle has started.
CPU/PCI Priority Queue PCI Priority Queue
CPU
PCI
PCI
PCI
REQ2#
REQ1#
REQ0#
PHLD#
REQ3#
Mode 1
Mode 0
PHLD#
055314
NOTES:
1. In the PCI Priority Queue, the last agent granted is always dropped to the bottom of the queue for the next arbitration
cycle, but the order of the chain is always preserved.
2. In the PCI Priority Queue, if PHLD is at the bottom of the queue, the upper PHLD slot is masked. This prevents back-to-
back grants, if other PCI requests are pending.
3. In the CPU/PCI priority Queue, the CPU is granted high priority status after 3 (mode 0) or 2 (mode 1) consecutive PCI
grants. If three (two) consecutive PCI grants have not been counted down, the CPU can be granted the bus as the low
priority agent.
Figure 12. Arbitration Priority Rotation
E82437VX (TVX) AND 82438VX (TDX)
65
PRELIMINARY
4.6.2. MULTI-TRANSACTION TIMER (MTT)
The priority chain algorithm has been enhanced by the Multi-Transaction Timer (MTT) mechanism. Once a PCI
agent is granted, the MTT is started. The timer counts down in PCI clocks from its preset value to zero. Until the
timer expires, that agent is promoted to being the highest priority PCI agent for the next grant event.
4.6.3. CPU POLICIES
The CPU never explicitly requests the bus. Instead, the arbiter grants the bus to the CPU when:
•the CPU is the highest priority
•PCI agents do not require main memory (peer-to-peer transfers or bus idle) and the PCI bus is not currently
locked by a PCI master
When the CPU is granted as highest priority, the MLT timer is used to guarantee a minimum amount of system
resources to the CPU before another requesting PCI agent is granted. An AHOLD mechanism controls granting
the bus to the CPU.
4.7. Shared Memory Buffer Architecture
The TVX provides a shared memory buffer interface allowing the frame buffer of an on-board video controller to
be implemented using main memory instead of dedicated video memory. The TVX contains a two-wire, dual
priority arbitration scheme. Both the TVX and graphics controller (GC) drive the system DRAM interface directly.
DRAM address, data, and control signals are alternately driven and tri-stated in a controlled manner. When the
SMB area is enabled (via the SMBCR Register), the TVX PCI REQ3#/GNT3# pin pair are used as the SMB
MREQ#/MGNT# arbitration signals.
The shared memory buffer resides in a programmable CPU address range between the value programmed in
the SMBSA Register and the top-of-memory. The top of main memory is the highest address indicated by the
DRAM Row boundary 4 (DRB4) Register.
NOTE
SDRAM is not supported in the row the graphics controller uses in a SMBA configuration. SDRAM with
EDO is supported in a SMBA configuration, if SDRAM is not populated in the SMBA DRAM row.
4.7.1. SMB SYSTEM MEMORY MAPPING
The Intel 430VX PCIset supports systems with either four rows of memory or five rows of memory (refer to the
DRAM Interface section). In a 4-row system, row 3 is always populated and is located at the top of the physical
address map. In a 5-row system, row 4 is always populated and located at the top of the physical address map.
Figure 13 illustrates the physical address map for a system with four DRAM rows:
82437VX (TVX) AND 82438VX (TDX) E
66
PRELIMINARY
Frame Buffer Aperture
Row 3
Top of System
Memory
Row 0
Increasing
Physical
Address
053315
Figure 13. Physical Address Map for 4 Rows of DRAM
The top of system memory (Figure 14) refers to the top of memory reported to the operating system; memory
above this may never be allocated by the operating system. The frame buffer aperture is located above the top
of system memory.
Note that for systems with only 16 Mbytes of memory and that provide memory holes to support legacy devices
that are mapped between 14 Mbytes and 16 Mbytes, the frame buffer must be mapped below the memory hole.
The Intel 430VX PCIset supports 1-Mbyte or 2-Mbyte memory holes sizes. A 1-Mbyte hole is located from 15−16
Mbytes and 2-Mbyte holes is located from 14−16 Mbytes. For example, for a 1-Mbyte frame buffer and a
memory hole of 2 Mbtes, the frame buffer must be mapped from 13−14 Mbytes and the memory hole is from 14−
16 Mbytes. The top of system memory is 13 Mbytes. The GC obtains the size and location of SMB areas
through the BIOS services. The Intel 430VX PCIset also supports an option that allows overlap between the
SMB area and the memory hole. In this case, the Intel 430VX PCIset gives priority to the hole over the SMB
area. All the CPU cycles within the hole and SMB area are forwarded down to PCI. The physical addresses for
the hole should not be decoded by the GC on the PCI bus. The graphics driver should ensure that the CPU
generates addresses above the top of DRAM. Figure 14 illustrates the mapping of the memory hole with the
frame buffer mapped below the memory hole.
Frame Buffer Aperture
Top of System
Memory
0 MB
Increasing
Physical
Address
2 MB PCI Hole
13 MB
14 MB
16 MB
Top of DRAM
055316
Figure 14. Frame Buffer with PCI Hole Enabled with 16-Mbyte system DRAM
E82437VX (TVX) AND 82438VX (TDX)
67
PRELIMINARY
4.7.2. CPU ACCESS TO SMB
The CPU has access to the entire DRAM address space including SMB. The system DRAM within the SMB
address range is non-cacheable.
Note that the CPU may access the frame buffer via three different mechanisms. If the aperture is closed,
addresses to the FB aperture directly above top of system memory are forwarded to PCI by the TVX. If the
aperture is open, the TVX accesses the FB region above top of system memory directly. Bit 0 in the SMBCR
register is used to direct CPU accesses to the FB aperature to either PCI or main memory. Also, the relocatable
address range specified in the Device Independent Configuration Space can be used and the TVX forwards
these cycles to PCI.
4.7.3. PCI MASTER ACCESS TO SMB
When the SMB space is enabled (Aperture can be either open or close) the TVX never responds to PCI master
memory read/write cycles whose address fall above the top of main memory. Instead, these PCI cycles must be
serviced by the GC which also resides on the PCI bus as a PCI slave (i.e., the GC should assert DEVSEL# to
claim these cycles). The GC must map its frame buffer within the PCI address space at the same location it
resides within the CPU address space.
4.7.4. GC (GRAPHICS CONTROLLER) ACCESS TO SMB
When the SMB space is enabled, the GC may have access to the frame buffer for screen refresh and draw
operations directly at the DRAM interface through the MREQ#/MGNT# handshake mechanism. GC buffer-to-
DRAM address bit mapping for the shared DRAM row(s) must match the TVX implementation for a one-to-one
linear correspondence. The address mapping for the TVX is provided in the DRAM Interface Section.
82437VX (TVX) AND 82438VX (TDX) E
68
PRELIMINARY
4.7.5. SMB INTERFACE
Figure 15 shows the interconnection of TVX, GC, and memory using the EDO/SPM signals, in a 4-row system.
In a 5-row system, the graphics controller would be connected to row 5 (i.e., RAS4#). Refer to the DRAM
Organization section, in the DRAM Interface section, for additional details on a 5-row implementation.
GCWE#
GCMAddr
CAS[7:0]#
RAS3#
GC
MD[63:0]
Optional
Buffers
DRAM
Row 3
WEB#
RAS3#
MD[63:0]
DRAM
Row 2
WEB#
RAS2#
MD[63:0]
DRAM
Row 1
WEA#
RAS1#
MD[63:0]
DRAM
Row 0
WEA#
RAS0#
MD[63:0]
MREQ#
MGNT#
TVX
HCLK
PCI
Bus
CAS[7:0]#
MAddr
WE[A:B]#
RAS[3:0]#
Host CPU
Interface
To TDX x 2
MD[63:0]
055317
NOTES:
1. The rows of memory shown above represent logical rows. In an actual system, double sided SIMMs would be used. There
are two SIMMs per row for a 64-bit memory interface.
2. If double-sided SIMM sockets are used on the motherboard, then the RAS3# from the GC and TVX should be tied to the
first bank of the double-sided SIMM and RAS2# from the TVX should be tied to the second bank of the double-sided
SIMM. This is required for supporting single-sided SIMMs populated in the double-sided SIMM sockets.
3. TVX provides two copies of WE# signal. Therefore, the GCWE# signal from GC needs to drive only two rows of memory.
4. The host bus clock is used by both the TVX and the GC to ensure correct operation of the synchronous memory bus
arbitration protocol.
5. The buffers shown on GCMAddr and GCWE# are optional for loading purposes. The use of these buffers will degrade
performance, take up real estate on the mother board, and also add to the system cost. The use of these external buffers
is not recommended. TVX is designed to work with the memory array with integrated buffers.
Figure 15. SMB Interconnect for EDO/SPM Signals
E82437VX (TVX) AND 82438VX (TDX)
69
PRELIMINARY
Figure 16 shows the interconnection of TVX, GC, and memory using the SDRAM signals, in a 4-row system. In
a 5-row system, the graphics controller would be connected to row 5 (i.e. CS5#). Refer to the DRAM
Organization section, in the DRAM Interface section, for additional details on a 5-row implementation.
GCWE#
GCMAddr
DQM[7:0]#
CS3#
GC
MD[63:0]
Optional
Buffers
DRAM
Row 3
SRASB#,
SCASB#
WEB#
RAS3#
MD[63:0]
DRAM
Row 2
RAS2#
MD[63:0]
DRAM
Row 1
RAS1#
MD[63:0]
DRAM
Row 0
RAS0#
MD[63:0]
MREQ#
MGNT#
TVX
HCLK
PCI
Bus
DQM[7:0]#
MAddr
WE[A:B]#
CS[3:0]#
Host CPU
Interface
To TDX x 2
MD[63:0]
SRASB#,
SCASB#
WEB#
SRASA#,
SCASA#
WEA#
SRASA#,
SCASA#
WEA#
HCLK
HCLK
HCLK
HCLK
SCAS[A:B]#
SRAS[A:B]#
GCCAS#
GCRAS#
055318
NOTES:
1. The rows of memory shown above represent logical rows. In an actual system, double-sided DIMMs would be used.
2. If a double-sided DIMM sockets are used on the motherboard, then the CS3# from the GC and TVX should be tied to the
first bank of the double-sided DIMM and CS2# from the TVX should be tied to the second bank of the double-sided DIMM.
This is required for supporting single-sided DIMMs populated in the double-sided DIMM sockets.
3. TVX provides two copies of WE#, SRAS#, and SCAS# signals. Therefore, the GCWE#, GCSRAS#, and GCSCAS#
signals from GC needs to drive only two rows of memory.
4. The host bus clock is used by the TVX, the GC, and the SDRAM to ensure correct operation of the synchronous memory
bus arbitration protocol. The memory interface of the GC must run at the same frequency as the host clock frequency.
5. The buffers shown on GCCAS#, GCRAS#, GCMAddr and GCWE# are optional for loading purposes. The use of these
buffers will degrade performance, take up real estate on the mother board, and also add to the system cost. The use of
these external buffers is not recommended. TVX is designed to work with the memory array with integrated buffers.
(version buffered of the DIMMs is also not supported.)
Figure 16. SMB Interconnect for SDRAM Signals
82437VX (TVX) AND 82438VX (TDX) E
70
PRELIMINARY
4.7.6. SMBA ARBITRATION
The TVX SMBA arbitration mechanism uses a two-wire MREQ#/MGNT# protocol. The protocol signaling is
synchronous to HCLK. The SMBA arbitration mechanism provides the following features:
•The TVX is the default owner (i.e., parked on the main memory DRAM bus)
•The GC has the ability to signal a low priority request or a high priority request using the MREQ# signal. The
low priority request allows the system performance impact to be minimized, preventing unnecessary system
resource stalls during non-critical GC accesses (i.e., performing draw operations or topping off of the GC’s
refresh FIFO). The high priority request provides a guaranteed latency for critical operations and should be
used to prevent GC FIFO underruns or when visible graphics degradation occurs if the bus can not be
acquired immediately (i.e., to prevent visible artifacts if the drawing engine is starving due to not being able
to acquire the bus via a low priority request).
•The maximum latency seen by the graphics controller when asserting a high priority request is
approximately 400ns (This is programmable. Refer to the TVX-to-GC Latency Control Management
Section). The maximum latency seen for a low priority request is non-determinant since the TVX only
responds to a low priority request if there are no CPU, PCI, or high priority refresh requests pending.
4.7.7. TVX-TO-GC LATENCY CONTROL MANAGEMENT
It is imperative that the memory controller (TVX) and graphics controller be sensitive to each other’s latency and
bandwidth requirements since overall system performance, visible artifacts, and/or limited functionality may
result. This is an area where differentiation can play a large role in improving overall system capabilities and
performance.
The Intel 430VX PCIset has been designed to make efficient use of the memory bus when it is the owner and to
release the bus to the graphics controller with minimal latency. To help control this latency the TVX provides a
programmable timer (GCLT Register) that limits the number of non-GC DRAM cycles that can be completed
after a high priority GC request has been sampled on the MREQ# pin. This timer maximizes efficient use of the
TVX buffering so that non-GC cycles that involve main memory can continue on during GC DRAM accesses,
using up available internal buffering.
The programmable GC latency timer is used to determine when to give up DRAM (on PCI reads and CPU/PCI
writes) when a high priority GC request is asserted. For PCI reads, the idea is to give up the bus (only after the
GC high priority request has been sampled active), as soon as DRAM is not being used efficiently (i.e., PCI Pre-
fetch buffers are full), or the GC latency timer times-out, which ever occurs first. For PCI-to-DRAM single writes,
the DRAM is relinquished when the GC latency timer times-out (after the current cycle has been written to
DRAM from the DRAM write buffers), or the DRAM write buffers become empty, whichever occurs first. The GC
latency timer is used for PCI-to-DRAM reads (burst or single), PCI-to-DRAM single writes (Note: CPU/PCI burst
writes always break on line boundaries when a high priority GC request is active), and CPU-to-DRAM single
writes.
The GC latency timer begins counting when the GC high priority request is sampled active and a PCI master is
reading or writing to DRAM or the CPU is writing to DRAM. The timer can be programmed to 0, 4, 8, 12, 16, 20,
24, and 28 HCLKs (defaults to 12 for CPU/PCI writes and 16 for PCI reads). For PCI reads, the default value will
guarantee that at least a line will be read from DRAM before DRAM is given to the GC. For PCI/CPU single
writes, the idea is to allow four single writes (page hits) to drain from the DRAM Write Buffers before giving
DRAM to the GC. This potentially opens up the DRAM write buffers for posting when the GC has DRAM. The
default values for this timer are set to guarantee that the maximum latency seen by the graphics controller is
approximately 400ns. The programmability is provided for performance timing purposes. Programmability in the
TRDY timer (TPDT Register) has also been provided for performance timing purposes. The TRDY timer is used
to determine how long the Intel 430VX PCIset waits before disconnecting the PCI transfer.
E82437VX (TVX) AND 82438VX (TDX)
71
PRELIMINARY
4.8. Clock Generation and Distribution
The TVX and CPU should be clocked from one clock driver output to minimize skew between the CPU and TVX.
The TDXs should share another clock driver output.
4.8.1. RESET SEQUENCING
The TVX is asynchronously reset by the CPU RESET signal. After RESET is negated, the TVX resets the two
TDXs by driving HOE#=MOE#=POE#=1 and MADV#=0 for two HCLKs. The TVX changes HOE#, MOE#, POE#
and MADV# to their default value synchronous to HCLK.
Arbiter (Central Resource) Functions on Reset
The TVX arbiter includes support for PCI central resource functions. These functions include driving the
AD[31:0], C/BE[3:0]#, and PAR signals when no one is granted the PCI bus and the bus is idle. The TVX drives
0s on these signals during these times, plus during RESET.
82437VX (TVX) AND 82438VX (TDX) E
72
PRELIMINARY
5.0. PINOUT AND PACKAGE INFORMATION
5.1. TVX Pinout
15 7
158
159
160
161
162
163
164
165
166
167
168
169
17 0
171
172
17 3
174
175
176
17 7
178
179
180
181
18 2
18 3
184
185
186
187
188
189
190
191
192
19 3
194
195
196
197
198
199
200
201
202
203
20 4
20 5
206
207
208
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
1
2
3
4
5
7
6
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52 105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
VSS
AD0
TWE#
TIO0
TIO1
TIO2
TIO7
TIO6
TIO5
TIO4
TIO3
VSS
CADV#/CAA4
CADS#/CAA3
COE#
GWE#
BWE#/CGCS#
CCS#/CAB4
KRQAK/CAB3
MXS/CS4#
PHLDA#
GNT0#
GNT1#
GNT2#
GNT3#/MGNT#
VSS
VDD3
A23
A21
A24
A27
A22
A26
A25
A28
A31
A3
VSS
A30
A29
A4
A7
A6
A5
A8
A11
A10
A16
A17
A18
V19
VSS
AD22
C/B E3#
AD25
VDD3
V DD 3
AD31
AD30
AD29
AD28
AD27
AD26
AD24
AD23
VSS
VDD3
AD21
AD20
AD19
AD18
AD17
AD16
C/BE 2#
FRAME#
IRDY#
TRDY#
VSS
VDD3
DEVSE L#
STOP #
L OC K#
PAR
C/BE1#
AD15
AD14
AD13
AD12
AD11
AD10
VSS
VDD3
AD9
AD8
C/BE0#
AD7
AD6
AD5
AD4
AD3
AD2
AD1
RE FVDD5
VD D 3
VSS
PLINK15
MSTB1
PCMD0
PCMD1
HOE#
POE#
VSS
WEA#
WEB#
MA0
MA1
MA2
MA3
MA4
VDD 3
MA5
MA6
MA7
MA8
MA9
MA10
MA11
VDD3
VSS
MOE#
RAS2#/CS2#
RAS3#CS3#
RAS0#/CS0#
RAS1#/CS1#
CAS7#/DQM7
CAS3#/DQM3
CAS5#/DQM5
CAS1#/DQM1#
CAS4#/DQM4
CAS0#/DQM0
CAS6#/DQM6
CAS2#/DQM2
SRASB#/RAS4#
SRASA#
SCASB#
SCASA#
REQ3#/MREQ#
REQ2#
REQ0#
PHLD#
PCLKIN
VSS
VSS
MSTB0
MADV#
REQ1#
VDD3
VDD3
A20
A9
A12
A14
A13
A15
HLOCK#
M /IO#
CACHE#
KEN#/INV
AHOLD
BRDY #
NA#
BOFF#
EADS#
ADS#
HITM#
W/R#
SMIACT#
VSS
HCLKIN
C PU RST
VDD3
VSS
BE 0#
BE3#
BE 5#
BE 6#
BE 7#
PLINK 0
PLINK1
PLINK2
PLINK4
PLINK5
PLINK6
PLINK7
PLINK8
PLINK10
PLINK13
PLINK14
VDD3
VDD3
PLINK11
PLINK12
D/C#
BE 1#
BE 2#
BE 4#
PLINK9
PLINK 3
055319
Figure 17. TVX Pinout
E82437VX (TVX) AND 82438VX (TDX)
73
PRELIMINARY
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
A3 37 I/O
A4 40 I/O
A5 43 I/O
A6 42 I/O
A7 41 I/O
A8 44 I/O
A9 56 I/O
A10 46 I/O
A11 45 I/O
A12 57 I/O
A13 59 I/O
A14 58 I/O
A15 60 I/O
A16 47 I/O
A17 48 I/O
A18 49 I/O
A19 50 I/O
A20 55 I/O
A21 29 I/O
A22 32 I/O
A23 28 I/O
A24 30 I/O
A25 34 I/O
A26 33 I/O
A27 31 I/O
A28 35 I/O
A29 39 I/O
A30 38 I/O
A31 36 I/O
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
AD0 2I/O
AD01 206 I/O
AD02 205 I/O
AD03 204 I/O
AD04 203 I/O
AD05 202 I/O
AD06 201 I/O
AD07 200 I/O
AD08 198 I/O
AD09 197 I/O
AD10 194 I/O
AD11 193 I/O
AD12 192 I/O
AD13 191 I/O
AD14 190 I/O
AD15 189 I/O
AD16 177 I/O
AD17 176 I/O
AD18 175 I/O
AD19 174 I/O
AD20 173 I/O
AD21 172 I/O
AD22 171 I/O
AD23 168 I/O
AD24 166 I/O
AD25 165 I/O
AD26 164 I/O
AD27 163 I/O
AD28 162 I/O
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
AD29 161 I/O
AD30 160 I/O
AD31 159 I/O
ADS# 70 I
AHOLD 65 O
BE0# 80 I
BE1# 81 I
BE2# 82 I
BE3# 83 I
BE4# 84 I
BE5# 85 I
BE6# 86 I
BE7# 87 I
BOFF# 68 O
BRDY# 66 O
BWE#/CGCS# 17 O
C/BE0# 199 I/O
C/BE1# 188 I/O
C/BE2# 178 I/O
C/BE3# 167 I/O
CACHE# 63 I
CADS#/CAA3 14 O
CADV#/CAA4 13 O
CAS0#/DQM0 142 O
CAS1#/DQM1 140 O
CAS2#/DQM2 144 O
CAS3#/DQM3 138 O
CAS4#/DQM4 141 O
CAS5#/DQM5 139 O
82437VX (TVX) AND 82438VX (TDX) E
74
PRELIMINARY
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
CAS6#/DQM6 143 O
CAS7#/DQM7 137 O
CCS#/CAB4 18 O
COE# 15 O
CPURST 77 I
D/C# 71 I
DEVSEL# 184 I/O
EADS# 69 O
FRAME# 179 I/O
GNT0# 22 O
GNT1# 23 O
GNT2# 24 O
GNT3#/MGNT# 25 O
GWE# 16 O
HCLKIN 76 I
HITM# 72 I
HLOCK# 61 I
HOE# 112 O
IRDY# 180 I/O
KEN#/INV 64 O
KRQAK/CAB3 19 I/O
LOCK# 186 I/O
M/IO# 62 I
MA0 117 O
MA1 118 O
MA2 119 O
MA3 120 O
MA4 121 O
MA5 123 O
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
MA6 124 O
MA7 125 O
MA8 126 O
MA09 127 O
MA10 128 O
MA11 129 O
MADV# 109 O
MOE# 132 O
MSTB0 106 O
MSTB1 108 O
MXS/CS4# 20 O
NA# 67 O
PAR 187 I/O
PCLKIN 154 I
PCMD0 110 O
PCMD1 111 O
PHLD# 153 I
PHLDA# 21 O
PLINK00 88 I/O
PLINK1 89 I/O
PLINK2 90 I/O
PLINK3 91 I/O
PLINK4 92 I/O
PLINK5 93 I/O
PLINK6 94 I/O
PLINK7 95 I/O
PLINK8 96 I/O
PLINK9 97 I/O
PLINK10 98 I/O
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
PLINK11 99 I/O
PLINK12 100 I/O
PLINK13 101 I/O
PLINK14 102 I/O
PLINK15 107 I/O
POE# 113 O
RAS0#/CS0# 135 O
RAS1#/CS1# 136 O
RAS2#/CS2# 133 O
RAS3#/CS3# 134 O
REFVDD5 207
REQ0# 152 I
REQ1# 151 I
REQ2# 150 I
REQ3#/MREQ# 149 I
SCASA# 148 O
SCASB# 147 O
SMIACT# 74 I
SRASA# 146 O
SRASB#/RAS4# 145 O
STOP# 185 I/O
TIO0 4I/O
TIO1 5I/O
TIO2 6I/O
TIO3 11 I/O
TIO4 10 I/O
TIO5 9I/O
TIO6 8I/O
TIO7 7I/O
E82437VX (TVX) AND 82438VX (TDX)
75
PRELIMINARY
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
TRDY# 181 I/O
TWE# 3O
VDD3 27
VDD3 53
VDD3 54
VDD3 78
VDD3 103
VDD3 104
VDD3 122
VDD3 130
VDD3 157
VDD3 158
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
VDD3 170
VDD3 183
VDD3 196
VDD3 208
VSS 1
VSS 12
VSS 26
VSS 51
VSS 52
VSS 75
VSS 79
VSS 105
Table 18. TVX Alphabetical
Pin List
Name
Pin
Type
VSS 114
VSS 131
VSS 155
VSS 156
VSS 169
VSS 182
VSS 195
W/R# 73 I
WEA# 115 O
WEB# 116 O
82437VX (TVX) AND 82438VX (TDX) E
76
PRELIMINARY
5.2. TDX Pinout
100
99
98
97
96
95
94
93
92
91
90
89
88
87
86
85
84
83
82
81
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
80
79
78
77
76
75
74
73
72
71
70
69
68
67
66
65
64
63
62
61
60
59
58
57
56
55
54
53
52
51
V SS
HD3
HD4
HD5
HD6
VDD3
HD7
HD8
HD9
HD10
HD11
HD12
HD13
HD14
HD15
HD16
HD17
HD18
V SS
HD 19
HD 20
HD 21
HD 22
HD 23
HD 24
HD 25
VD D3
V SS
VDD3
HCLK
HD29
HD28
HD27
HD26
VSS
HD30
HD31
MD24
M D8
M D16
M D0
M D25
M D9
M D17
M D1
MD26
M D10
VDD3
VS S
M D18
HD2
HD1
HD0
PLINK0
PLINK1
PLINK2
PLINK3
PLINK4
PLINK5
PLINK6
PLINK7
M ST B0
M ST B1
VS S
VDD3
M AD V#
PC M D0
PCM D1
HO E#
POE#
MXS
VDD3
VSS
MOE#
VDD3
MD15
MD31
MD7
MD23
MD14
MD30
MD6
MD22
MD13
MD29
VSS
VDD3
MD5
MD21
MD4
MD20
MD12
MD28
MD3
MD19
MD11
MD27
VSS
REFVDD5
MD2
055320
Figure 18. TDX Pinout
E82437VX (TVX) AND 82438VX (TDX)
77
PRELIMINARY
Table 19. TDX Alphabetical
Pin List
Name
Pin
Type
HCLK 30 I
HD0 98 I/O
HD1 99 I/O
HD2 100 I/O
HD3 3I/O
HD4 4I/O
HD5 5I/O
HD6 6I/O
HD7 7I/O
HD8 8I/O
HD9 9I/O
HD10 10 I/O
HD11 11 I/O
HD12 12 I/O
HD13 13 I/O
HD14 14 I/O
HD15 15 I/O
HD16 16 I/O
HD17 17 I/O
HD18 18 I/O
HD19 20 I/O
HD20 21 I/O
HD21 22 I/O
HD22 23 I/O
HD23 24 I/O
HD24 25 I/O
HD25 26 I/O
HD26 32 I/O
HD27 33 I/O
Table 19. TDX Alphabetical
Pin List
Name
Pin
Type
HD28 34 I/O
HD29 35 I/O
HD30 36 I/O
HD31 37 I/O
HOE# 82 I
MADV# 85 I
MD0 41 I/O
MD1 45 I/O
MD2 51 I/O
MD3 57 I/O
MD4 61 I/O
MD5 63 I/O
MD6 69 I/O
MD7 73 I/O
MD8 39 I/O
MD9 43 I/O
MD10 47 I/O
MD11 55 I/O
MD12 59 I/O
MD13 67 I/O
MD14 71 I/O
MD15 75 I/O
MD16 40 I/O
MD17 44 I/O
MD18 50 I/O
MD19 56 I/O
MD20 60 I/O
MD21 62 I/O
MD22 68 I/O
Table 19. TDX Alphabetical
Pin List
Name
Pin
Type
MD23 72 I/O
MD24 38 I/O
MD25 42 I/O
MD26 46 I/O
MD27 54 I/O
MD28 58 I/O
MD29 66 I/O
MD30 70 I/O
MD31 74 I/O
MOE# 77 I
MSTB0 89 I
MSTB1 88 I
MXS 80 I
PCMD0 84 I
PCMD1 83 I
PLINK0 97 I/O
PLINK1 96 I/O
PLINK2 95 I/O
PLINK3 94 I/O
PLINK4 93 I/O
PLINK5 92 I/O
PLINK6 91 I/O
PLINK7 90 I/O
POE# 81 I
REFVDD5 52
VDD3 2
VDD3 27
VDD3 29
VDD3 48
82437VX (TVX) AND 82438VX (TDX) E
78
PRELIMINARY
Table 19. TDX Alphabetical
Pin List
Name
Pin
Type
VDD3 64
VDD3 76
VDD3 79
VDD3 86
VSS 1
Table 19. TDX Alphabetical
Pin List
Name
Pin
Type
VSS 19
VSS 28
VSS 31
VSS 49
VSS 53
Table 19. TDX Alphabetical
Pin List
Name
Pin
Type
VSS 65
VSS 78
VSS 87
E82437VX (TVX) AND 82438VX (TDX)
79
PRELIMINARY
5.3. TVX Package Dimensions
A
D
D1
L1
b
Units: m
m
152
53
104
105
156
157
208
TA1
Y
* Note* Height Measurements same
as Width Measurements
e1
C
055321
Figure 19. TVX Package Dimensions (208-Pin QFP)
Table
20
. TVX Package Dimensions (208-Pin QFP)
Symbol
Description
Value (mm)
A
Seating Height
4.25 (max)
A1 Stand-off
0.05 (min); 0.40 (max)
b
Lead Width
0.2
±
0.10
C
Lead Thickness
0.15 +0.1/-0.05
D
Package Length and Width, including pins
30.6
±
0.4
D1
Package Length and Width, excluding pins
28
±
0.2
e1
Linear Lead Pitch
0.5
±
0.1
Y
Lead Coplanarity
0.08 (max)
L1
Foot Length
0.5
±
0.2
T
Lead Angle
0°
- 10
°
82437VX (TVX) AND 82438VX (TDX) E
80
PRELIMINARY
5.4. TDX Package Dimensions
A
D
D1
L1
b
Units: mm
1
TA1
Y
e1
C
E
E1
055322
Figure 20. TDX Package Dimensions (100-Pin QFP)
Table
21
. TDX Package Dimensions (100-Pin QFP)
Symbol
Description
Value (mm)
A
Seating Height
3.3 (max)
A1 Stand-off
0.0 (min); 0.50 (max)
b
Lead Width
0.3
±
0.10
C
Lead Thickness
0.15 +0.1/-0.05
D
Package Width, including pins
17.9
±
0.4
D1
Package Width, excluding pins
14
±
0.2
E
Package Length, including pins
23.9
±
0.4
E1
Package Length, excluding pins
20
±
0.2
e1
Linear Lead Pitch
0.65
±
0.12
Y
Lead Coplanarity
0.1 (max)
L1
Foot Length
0.8
±
0.2
T
Lead Angle
0o
- 10
o
E82437VX (TVX) AND 82438VX (TDX)
81
PRELIMINARY
6.0. TESTABILITY
6.1. 82437VX TVX Testability
The test modes of TVX are summarized briefly in the following table and are described in detail in the later
sections.
The test modes are decoded from REQ[3:0]# and qualified with the CPURST pin. TVX test mode selection is
asynchronous. These signals need to remain in their respective states for the duration of the test mode
Table
22
. TVX Test Mode Summary
Test Mode
Detection of Mode
Description
NAND Tree
CPURST = ‘1’ AND REQ[2:0]# =
“000”
Enable NAND chain
IDCODE A
CPURST = ‘1’ AND REQ[2:0]# =
“110”
Drive out Revision ID and Device ID on AD lines
IDCODE B
CPURST = ‘1’ AND REQ[2:0]# =
“010”
Drive out Manufacturing ID on AD lines
TRISTATE
MODE
CPURST = ‘1’ AND REQ[2:0]# =
“100”
Float all outputs and disable pullups, pull downs
6.1.1. TVX NAND TREE MODE
Two NAND trees are provided in the TVX for Automated Test Equipment (ATE) board level testing. The NAND
trees allow the tester to test the connectivity of each of the TVX signal pins. While in NAND tree mode, all TVX
drivers except for GNT2# and GNT1# are tri-stated. The NAND tree outputs are on GNT2# and GNT1#. NAND
chain testing should be performed at 1 MHz (1000 ns ATE tester period). Allow 1000ns for the input signals to
propagate to the NAND tree outputs.
NAND chain #1 begins at MA8 and is routed counter-clockwise around the chip to its output on GNT1#. NAND
chain #2 begins at MA7 and is routed clockwise around the chip to its output on GNT2#. The NAND chain skips
CPURST, REQ2#, REQ1#, and REQ0# which are used to detect the NAND tree test mode.
Enter the NAND chain mode by driving CPURST to 1 and REQ[2:0]# = 000. To perform the NAND tree test after
entering this mode, all pins included in the NAND trees should be driven to 1. Beginning at the tree output
nearest the respective observability points (i.e. start with GNT0# and GNT3#), walk a zero toward the end of
each chain. As pins are toggled, the resulting ripple is observed on the corresponding NAND tree’s output. Since
the NAND tree test mode is selected asynchronously, REQ[2:0]# and CPURST must be stable for the duration
of the test.
82437VX (TVX) AND 82438VX (TDX) E
82
PRELIMINARY
Table
23
. TVX NAND TREE ONE
Tree
output
Pin #
Pin Name
Comments
23 GNT1#
GNT1# is output
of NAND chain #1
1 22 GNT0#
Last input to
chain. This pin is
logically closest to
GNT1#, output of
chain 1.
2 21 PHLDA#
3 20
MXS/
CS4#
4 19
KRQAK
CAB3
5 18
CCS#/
CAB4
6 17
BWE#/
CGCS#
7 16 GWE#
8 15 COE#
9 14
CADS#/
CAA3
10 13
CADV#/
CAA4
11 11 TIO3
12 10 TIO4
13 9 TIO5
14 8 TIO6
15 7 TIO7
16 6 TIO2
17 5 TIO1
18 4 TIO0
19 3 TWE#
20 2 AD0
21 206 AD1
Table
23
. TVX NAND TREE ONE
Tree
output
Pin #
Pin Name
Comments
22 205 AD2
23 204 AD3
24 203 AD4
25 202 AD5
26 201 AD6
27 200 AD7
28 199 C/BE0#
29 198 AD8
30 197 AD9
31 194 AD10
32 193 AD11
33 192 AD12
34 191 AD13
35 190 AD14
36 189 AD15
37 188 C/BE1#
38 187 PAR
39 186 LOCK#
40 185 STOP#
41 184 DEVSEL#
42 181 TRDY#
43 180 IRDY#
44 179 FRAME#
45 178 C/BE2#
46 177 AD16
47 176 AD17
48 175 AD18
49 174 AD19
50 173 AD20
E82437VX (TVX) AND 82438VX (TDX)
83
PRELIMINARY
Table
23
. TVX NAND TREE ONE
Tree
output
Pin #
Pin Name
Comments
51 172 AD21
52 171 AD22
53 168 AD23
54 167 C/BE3#
55 166 AD24
56 165 AD25
57 164 AD26
58 163 AD27
59 162 AD28
60 161 AD29
61 160 AD30
62 159 AD31
63 154 PCLKIN
64 153 PHLD#
65 149 REQ3#
66 148 SCASA#
67 147 SCASB#
68 146 SRASA#
69 145
SRASB#/
RAS4#
70 144
CAS2#/
DQM2
71 143
CAS6#/
DQM6
72 142
CAS0#/
Table
23
. TVX NAND TREE ONE
Tree
output
Pin #
Pin Name
Comments
DQM0
73 141
CAS4#/
DQM4
74 140
CAS1#/
DQM1
75 139
CAS5#/
DQM5
76 138
CAS3#/
DQM3
77 137
CAS7#/
DQM7
78 136
RAS1#/
CS1#
79 135
RAS0#/
CS0#
80 134
RAS3#/
CS3#
81 133
RAS2#/
CS2#
82 132 MOE#
83 129 MA11
84 128 MA10
85 127 MA9
86 126 MA8
MA8 is input to
NAND chain #1.
First input in the
chain, logically
farthest from
output point at
GNT1#
82437VX (TVX) AND 82438VX (TDX) E
84
PRELIMINARY
Table
24
. TVX NAND TREE TWO
Tree
output
Pin #
Pin Name
Comments
24 GNT2#
GNT2# is the
output of NAND
chain #2
1 25
GNT3#/
MGNT#
Last input to
chain. This pin is
logically closest
to GNT2#, output
of chain 2.
2 28 A23
3 29 A21
4 30 A24
5 31 A27
6 32 A22
7 33 A26
8 34 A25
9 35 A28
10 36 A31
11 37 A3
12 38 A30
13 39 A29
14 40 A4
15 41 A7
16 42 A6
17 43 A5
18 44 A8
19 45 A11
20 46 A10
21 47 A16
22 48 A17
23 49 A18
24 50 A19
Table
24
. TVX NAND TREE TWO
Tree
output
Pin #
Pin Name
Comments
25 55 A20
26 56 A9
27 57 A12
28 58 A14
29 59 A13
30 60 A15
31 61 HLOCK#
32 62 M/IO#
33 63 CACHE#
34 64 KEN#/INV
35 65 AHOLD
36 66 BRDY#
37 67 NA#
38 68 BOFF#
39 69 EADS#
40 70 ADS#
41 71 D/C#
42 72 HITM#
43 73 W/R#
44 74 SMIACT#
45 76 HCLKIN
46 80 BE0#
47 81 BE1#
48 82 BE2#
49 83 BE3#
50 84 BE4#
51 85 BE5#
52 86 BE6#
53 87 BE7#
E82437VX (TVX) AND 82438VX (TDX)
85
PRELIMINARY
Table
24
. TVX NAND TREE TWO
Tree
output
Pin #
Pin Name
Comments
54 88 PLINK0
55 89 PLINK1
56 90 PLINK2
57 91 PLINK3
58 92 PLINK4
59 93 PLINK5
60 94 PLINK6
61 95 PLINK7
62 96 PLINK8
63 97 PLINK9
64 98 PLINK10
65 99 PLINK11
66 100 PLINK12
67 101 PLINK13
68 102 PLINK14
69 106 MSTB0#
70 107 PLINK15
71 108 MSTB1#
72 109 MADV#
Table
24
. TVX NAND TREE TWO
Tree
output
Pin #
Pin Name
Comments
73 110 PCMD0
74 111 PCMD1
75 112 HOE#
76 113 POE#
77 115 WEA#
78 116 WEB#
79 117 MA0
80 118 MA1
81 119 MA2
82 120 MA3
83 121 MA4
84 123 MA5
85 124 MA6
86 125 MA7
MA7 is input to
NAND chain #2.
First input in the
chain, logically
farthest from
output point at
GNT2#.
82437VX (TVX) AND 82438VX (TDX) E
86
PRELIMINARY
Figure 22 is a schematic of the NAND Tree 1 circuitry. Figure 23 is a schematic of the NAND Tree 2 circuitry.
MA8
MA9
GNT0#
GNT1#
NAND CHAIN
SELECT
GNT1#
055323
Figure 21. 82437VX NAND Tree Diagram
MA7
MA6
GNT3#
GNT2#
NAND CHAIN
SELECT
GNT2#
055324
Figure 22. 82437VX NAND Tree Diagram
E82437VX (TVX) AND 82438VX (TDX)
87
PRELIMINARY
6.1.2. TVX ID CODE EXTRACTION
The two ID Code test modes provide a simple method of checking information stored in the TVX’s read-only
configuration registers at 02−03h and 08h. Like the other test modes, ID CODE is entered when TVX detects
particular patterns on REQ[2:0]# with CPURST active. Selection of this mode is asynchronous; when REQ#
pattern changes or CPURST is negated, the ID CODE mode is exited.
Table
25
. TVX ID Code Test Modes
MODE
REQ[2:0]#
detect pattern
Action
ID CODE A
110
Device ID driven on AD[31:16] (contents config reg 02-03h). Revision ID
driven on AD[7:0] (contents config reg 08h)
ID CODE B
010
Manufacturing ID driven on AD[31:0]
6.1.3. TVX TRI-STATE MODE
Tri-state mode is entered when TVX detects 100 on REQ[2:0]# and CPURST is active. In this mode all TVX
output drivers are tri-stated and internal buffer pull-ups and pull-downs are disabled. Selection of this mode is
asynchronous; when REQ# pattern changes or CPURST is negated, the TRISTATE mode is exited.
6.2. 82438VX TDX Testability
For the TDX, test modes are selected following synchronous detection of the TDX reset combination on HOE#,
MOE#, POE#, and MADV#. Value on MSTB[1:0] selects the Test Mode from the test options available.
Table
26
. TDX Test Mode Summary
Test Mode
Detection of Mode
Description
NAND Tree
1. Detect RESET (i.e., two samples of HOE#, MOE#,
POE# = 111 and MADV# = 0, with MSTB[1:0] = 00.)
2. Hold HCLK = 0 during NAND chain toggle.
Enable NAND chain,
observe output on MD0
Manufacturing
ID Code
1. Detect RESET (i.e., two samples of HOE#, MOE#,
POE# = 111 and MADV# = 0, with MSTB[1:0] = 01.)
2. Perform walking one sequence from HD0 to HD19.
Drive out Manufacturing
ID and part revision on
MD0 in response to
sequence driven in on
HD[19:0]
TRISTATE
MODE
1. Detect RESET (i.e., two samples of HOE#, MOE#,
POE# = 111 and MADV# = 0.)
2. Negate MADV# to 1 while continuing to drive HOE#,
MOE# and POE# to 111. Main busses will remain in tri-
state mode.
Float all outputs
Pull-down
Disable
1. Detect RESET (i.e., two samples of HOE#, MOE#,
POE# = 111 and MADV# = 0, with MSTB[1:0] = 11.)
2. Pulldowns will remain disabled until another reset is
detected.
Disable pull-downs
82437VX (TVX) AND 82438VX (TDX) E
88
PRELIMINARY
6.2.1. TDX NAND TREE MODE
The TDX has a single NAND tree provided for board level connectivity testing. While in NAND tree mode, all
TDX pins except MD0 are tri-stated. The NAND tree output is observed on MD0 (pin 41). NAND chain testing
should be performed at 1 MHz (1000 ns ATE tester period). Allow 1000ns for the input signals to propagate to
the NAND tree outputs.
The NAND chain begins at HD3 and is routed counterclockwise around the part, ending at HD2. The NAND
chain skips HCLK. HCLK connectivity is verified by correctly detecting RESET and entering NAND chain mode.
After entering NAND chain mode, drive all NAND chain inputs, including HOE#, POE#, MOE#, and MADV#, to
1. Since this test mode is synchronously evoked, HCLK should be driven to 0 for the duration of the test.
Clocking the part causes the TDX to exit NAND tree test mode. Starting at HD2, walk a zero back through the
chain toward the origin at HD3. As each pin is toggled, the resultant ripple is observed on MD0.
Table
27
. TDX NAND Tree
Tree
Output
PIN
Pin Name
Comments
41 MD0
Output of NAND
Chain, test mode
observability point
100 HD2
Last input pin to
NAND chain,
logically closest to
chain output.
Output observed
on MD0
99 HD1
98 HD0
97 PLINK0
96 PLINK1
95 PLINK2
94 PLINK3
93 PLINK4
92 PLINK5
91 PLINK6
90 PLINK7
89 MSTB0
88 MSTB1
85 MADV#
84 PCMD0
83 PCMD1
Table
27
. TDX NAND Tree
Tree
Output
PIN
Pin Name
Comments
82 HOE#
81 POE#
80 MXS
77 MOE#
75 MD15
74 MD31
73 MD7
72 MD23
71 MD14
70 MD30
69 MD6
68 MD22
67 MD13
66 MD29
63 MD5
62 MD21
61 MD4
60 MD20
59 MD12
58 MD28
57 MD3
E82437VX (TVX) AND 82438VX (TDX)
89
PRELIMINARY
Table
27
. TDX NAND Tree
Tree
Output
PIN
Pin Name
Comments
56 MD19
55 MD11
54 MD27
51 MD2
50 MD18
47 MD10
46 MD26
45 MD1
44 MD17
43 MD9
42 MD25
40 MD16
39 MD8
38 MD24
37 HD31
36 HD30
35 HD29
34 HD28
33 HD27
32 HD26
26 HD25
25 HD24
24 HD23
23 HD22
Table
27
. TDX NAND Tree
Tree
Output
PIN
Pin Name
Comments
22 HD21
21 HD20
20 HD19
18 HD18
17 HD17
15 HD16
15 HD15
14 HD14
13 HD13
12 HD12
11 HD11
10 HD10
9HD9
8HD8
7HD7
6HD6
5HD5
4HD4
3HD3
Start of NAND
Chain. HD3 is at
the logical end of
the chain, furthest
from the output
point at MD0
82437VX (TVX) AND 82438VX (TDX) E
90
PRELIMINARY
Figure 23 is a schematic of the NAND Tree circuitry.
HD3
HD4
HD2
Normal MD0 Output
NAND CHAIN
SELECT
MD0
055325
Figure 23. 82438VX NAND-Tree Diagram
6.2.2. TDX ID CODE EXTRACTION
The TDX does not have any addressable register space. Revision identification and manufacturing ID are
obtainable only through use of a test mode. To enter this mode, MSTB[1:0] should be held to ‘01’ while resetting
the part. Apply 00000h to HD[19:0]. Walk a one, once per test cycle, ascending from HD0 to HD19. The
combined revision ID and manufacturing ID are sequenced out on MD0. The resultant binary stream is broken
down as follows:
Bit 19 bit 16 bit 15 bit 0
Revision ID
Manufacturing ID
The Revision ID (bits [19:16]) for TDX A-0 is 0h. The Manufacturing ID (bits [15:0]) will read 0F20h, identical to
the TVX.
6.2.3. TDX TRI-STATE MODE
The HD, MD, and PLINK bus buffer control is provided at the periphery through HOE#, MOE# and POE#
respectively. Following the normal reset sequence, where HOE#, MOE#, and POE#=111 and MADV#=0 for two
consecutive HCLK’s, MADV# can be negated to 1 and the TDX remains in tri-state mode. Normally, MSTB[1:0]
are driven to 00 during reset. If these pins are driven to 11 during reset, the pull down resistors on the HD and
MD busses are also disabled. The HD and MD pull downs can be re-enabled using the normal reset sequence
where MSTB[1:0] are 00.
