Type 2 Processor Complex

rf90952a.exe Reference Disk Type 2 Processor Complex v1.21 (zipped image)
rd9095a.exe Diagnostic Disk Type 1 - 4 Processor Complex v2.33 (zipped image)
41g8386.zip Reference Disk Type 2 Processor Complex v1.10 (zipped image only)
   Old version! Supports Model 90 XP 486 & 95 XP 486 with Type 2 Processor Complex.
41g8575.zip Reference Disk Model 90 XP 486 Type 2 Complex v1.02 (zipped image only)
   Old version! Supports only Model 90 XP 486 with Type 2 Processor Complex. Includes diagnostics.

192-100 IBM PS/2 486-25/50 Microprocessor Upgrade (CPU only!)
192-099 IBM PS/2 486DX2-50 Processor Upgrade

486 Interposers and Upgrades

Specifications
   Memory
   Features
Complex Origins
"H" and "L" Complex Identification
"H" / Upgrade 486SX 25 MHz
"L" / Upgrade 486DX2 50 MHz
Rear of "H" / "L"
J1 Jumper
U6 Variants and Kingston Turbochip
Running DX4-100 on L (Requires U23)
3172-002 "H" complex 93F1666
Overclock attempt
ODP vs. ODPR
ODP in Original Socket
In Defense of the L
System Firmware
   ROM Images
   Support for >1 GB Disks as IML Drive
Known Problems
   171 POST Errors
   Diskette Data Loss
   16-bit Busmasters

Japanese 486SX 25 MHz / 486DX2 50 MHz Complex
Japanese DX2-66 Upgrade Type A Complex (unique!)
Japanese DX2-66 Upgrade Type B Complex


Specifications

Memory

RAM

  • Min/Max on system board: 8 / 64 MB, 16 MB addressable by DMA
  • PS/2 70 ns parity 72-pin SIMMs, 85 ns SIMMs not supported

ROM

  • 128 KB EPROM

Cache

  • L1: 8 KB (486 SX, DX2)
  • L2: none (no socket for cache)

Features

  • No Level 2 cache socket on complex.
  • H models socketed for a 487 copro or a 486DX2 50 MHz upgrade chip.
  • High speed 25 MHz DMA - now synchronous with 486; 24-bit DMA. *
  • Faster bus arbitration (than Base 1) for better busmaster performance.
  • Memory controller supports both interleaved (higher performance pairs of SIMMs) and non-interleaved memory (allows single SIMMs).
  • 20 MB per second data transfer support (for MCA bus).

* - The same DMA controller chip is used in other systems that support 32-bit DMA (namely the Type 3 complex, 85 X planar, 85 K/N planar, and other. So is this a case of a 32-bit part being limited to 24-bit by the board and/or firmware implementation?


Complex Origins

The Type 2 complex was most likely designed by IBM Japan. This is supported by the following facts:

  • The core logic consists of Japanese gate arrays (QFP package)
  • The PCB itself looks very similar to the other boards that were designed by IBM Japan. Namely - solder trap pads for wave soldering (on the backside), silkscreen in the more descriptive "position/type" format, etc.
  • The firmware has support for the PS/55 Model 5560 planar (Planar ID F9FE).

"H" and "L" Complex Identification

Both H and L have no L2 cache socket or solder pads for one.

H models (92F0079) came with a 486SX-25 in U8 and have a second socket (U9) for a 487 copro or a 486DX2-50 ODP upgrade chip. They will accept a Turbochip if the DMA controller is the good one.

L models (92F0161) came with a 486DX2-50 in U8 and have only solder pads at the U9 location. They will take a ODPR or (with the good DMA controller) a Turbochip.


"H" / Upgrade 486SX 25 MHz 92F0079, FCC ID ANOIBM486SXB25 [P] [P] | [P] [P]

U5 (GA-J) 89F5724 Memory Controller
U6 (GA-M) 92F1428 DMA Controller
U7 07G0463 BIOS ROM Am27C010-155DC
U8 169-pin LIF CPU socket (486SX-25)
U9 487 socket
U10 (GA-T) 64F8781 Memory Data Buffer
U16 (GA-R) 89F5415 MCA Buffer/Controller?
Y1 50.0000 MHz osc (half can or SMD)

Note: There are two slightly different versions of the PCB. The most noticeable difference on the front side is the unpopulated position U38 - the older revision has pads for a square PLCC-20/28 package while the newer one has two rows of pads for a SOIC-16/20 package.


"L" / Upgrade 486DX2 50 MHz 92F0161, FCC ID ANOIBM486SXB25 [P]

U5 (GA-J) 89F5724 Memory Controller
U6 (GA-M) DMA Controller
U7 BIOS ROM Am27C010-155DC
U8 169-pin LIF CPU socket (486DX2-50)
U9 Solder pads for 487 socket
U10 (GA-T) 64F8781 Memory Data Buffer
U16 (GA-R) 89F5415 MCA Buffer/Controller?
Y1 50.0000 MHz osc (half can or SMD)


Rear of "H" / "L"

U23 74F08 (SOIC-14) or pads for SOIC-20 (see note below and also info about running DX4).

Note: There are two slightly different versions of the PCB. On the back side there is one significant change - on the older version position U23 is unpopulated (SOIC-20) on the newer one there's 74F08 (SOIC-14) in the same place.


J1 Jumper

Right below the CPU socket there are pads for a 2-pin jumper marked as "(J1)". What is it for?


U6 Variants and Kingston Turbochip

Some Type 2 (92F0079) suffer from an "incompetent DMA-chip", which is P/N 92F1428 at position U6 on the card. Working Type 2s 92F0079 have a DMA-chip P/N 10G7808 at U6. If U6 is 10G7808 Kingston Turbochip should work fine. The earlier Type 2 use the 92F1428 - which ends the experiment in odd results (permanent I9990044, 605, 165 errors, and inability to read from FDD). Known U6 variants:

  • P/N 92F1428 (Bad)
  • P/N 10G7808 (Good)
  • P/N 10G4714 (Possibly Good)

Running DX4-100 on L (Requires U23)

Jay Bodkin said (edited):
   I got a bizarre thing happening on my 8595 Type 2. I upgraded the CPU from an Intel 486 DX2-50 to 486 DX4 -100. (I got the U6 P/N 10G4714). The machine boots with the reference disk and it detects that the processor complex has changed. The front panel has the usual code to go with that. Then it asks to perform automatic configuration, to which I replied y. Next, it complains that there are still unresolved configuration errors, and wants a reboot. After another reboot the front panel displaying 9600-8N1 and refuses to boot further.

Peter Wendt replies:
   OK, so at least you're not in danger toasting the CPU. (Tastes pretty bad anyway - toasted or raw) FWIW: I ran the Kingston Turbochip-133 at 100MHz on various 25MHz Type 2 platforms, but all of them had the "competent" DMA chip 10G7808 at U6/GA-M *and* had the latest platform BIOS 41G9361.

All "L" complexes have U23.
Complex PCB is 06G6870, BIOS 07G0463, DMA 10G7808, sticker 48G8043
Complex PCB is 06G6870, BIOS 41G9361, DMA 10G4714, sticker 06G7009

"H" complexes with 48G8042 on sticker below CPU have U23
Complex PCB is 06G6870, BIOS 07G0463, DMA 10G7808

"H" complexes with 07G3152 on sticker below CPU do NOT have U23
Complex PCB is 06G4561, BIOS 07G0463, DMA 92F1428

The presence of a 74F08 chip at position U23 (on the solder side of the board) seems also be a critical point: if it is there the board is made in 1992, earlier 1991-manufactured board lack this chip - and failed to work with anything faster than a DX2-25/50 (Or a DX2-33/66, which will run as a 25/50 there only anyway). DX4s or Evergreen, Kingston and such like based on a quad-clocked AMD 5x86 failed with no exception on the earlier ones.

There should be a silk-screened "Date of MFG" number somewhere on the platform, like e.g. 2092A0700 - which means 20th week 1992 (first 4 digits). If yours is a 1991 platform or early 1992 without the U23 chip - forget about it. On these boards I had a failure-rate of 100% (about 20 out of 20...).

> Interestingly, putting back the 486 DX-2 50 doesn't work now either. It still gets stuck on the 96-8N1, even after removing the battery for a couple of minutes!

IBM wrote in the HMM "Remove the battery, then wait 5 minutes..." (HMM, September 1993, P/N 71G9316, page 300, "96 8N1 Error Message") but empirical research showed that this doesn't work fairly often.

Shortening the battery connectors (with the battery removed of course) *and* toggling the startup password jumper seems to be the only fast cure. There seems to be a board logic, that ANDs the two conditions after a power-on and deletes the entire setup from the CMOS.

At least during the "hot phase" when the Mod. 90s and 95s can be found in larger amounts at the customers this procedure was the only one that worked in a sufficiently fast way. And I had quite a lot 95s under service... ;-) (Some are mine now)


3172-002 93F1666

The 3172-002 uses a 95 single serial planar and an "H" complex with a 3172 flavor complex BIOS (93F1667). This particular 3172-002 complex is the later version with a 10G7808 DMA controller and U23 on the back. They will happily accept the 41G9361 BIOS.


Overclock attempt

Swapping a 66.667 MHz osc in for the 50 MHz one will result in a 00010200 code in the upper left hand of the screen and a system hang. The Y1 Oscillator drives both the data bus and the DMA controller. 0001 02XX ROM checksum or timer error.


ODP vs. ODPR

The 169-pin is the ODP-version - substitutes a 487SX with the 486SX still in place. Important on boards with soldered CPU or a second socket.
The 168 pin is the ODPR-version, where the -R- stands for "Replace"... it replaces the 486SX / 486DX respectively and does not need the "SX-disable" pin therefore.

ODP in Original Socket

Dr. Jim, can the 169th pin be bent/removed and the CPU used?

Yep. The socket I just put into my P75 has an empty hole to allow that extra pin to pass. I've drilled similar holes in older 486 sockets myself. Carefully. BTW, I've been informed that it is NOT the SX disable pin, just a key pin. The SX disable is elsewhere.

Jose Duran:
   You are right, this pin is only a key pin without any signal. The SX disable pin is B14 (as far as I remember). I have installed an AMD DX5 on an upgrade socket with a voltage adapter. Since my CPU didn't have any SX disable signal on it. I had to solder a wire from pin B14 to ground, thus putting the SX into tristate. It works perfectly.


In Defense of the L

Actually Tony Ingenoso said:
   Austin DID do some LAN server performance modeling along these lines and determined that there WAS a gain to be had from a dual 386 machine as a 386 file server could become CPU bound. Once a 486DX-33 was fitted, even the heaviest traffic hitting the server wasn't enough to saturate the DX33 in a pure file server role.

My own casual observations of the lowly (and IMO brilliant) DX2-50 T2 complex validated Austin's results. With all NT4 Server's performance monitoring turned on I was never unable to saturate the stock CPU on a 9595-0LF in a pure file server role. It might run up to 80%, maybe even peak to 90%, but it never became saturated at 100%. This was with several other PC hammering it over a 16mb T/R LAN. Operating within its design purpose, doing what it was sold to do -- a stock T2 was is a word "sufficient" to the task in all respects, and remains so even today when run in a pure file server role.

The problem is that TODAY, we are asking these machines to take on more than they were intended to do initially. Modern "desktop" usage has a dramatically different use profile than pure file servers have, and it shows. GUI's are a big drag on performance, and the 95's placement of the video on the bus hurts quite a bit as well.


System Firmware (POST & BIOS)

Simplified firmware stored in EPROM. IML image required for the system to operate.

ROM Images

All ROMs are 27C010 or compatible (position U7).

07G0463 08 Aug 1991, rev. 0, Type 2 (early "H"), seen mostly with 92F1428 DMA Ctrl.
41G9361 03 Mar 1992, rev. 3, Type 2 ("H" or "L"), stock on Hs with 10G7808 DMA Ctrl.

Support for >1 GB Disks as IML Drive

Type 2 complexes require the combination of BIOS 41G9361 and SCSI BIOS 92F2244/45 in order to handle IML drives >1GB (new limit is 3.94GB). The upgrade BIOS incorporates the "Enhanced IML" which supports IML from a drive >1GB and "Search IML" which allows IML from a drive other than ID6. The SCSI BIOS 44/45 pair supports drives well over 8GB. (Ed. You OS may have other ideas...)

The old' complex ROMs' IML support somehow does not make (proper) use of "Enhanced IML", possibly due to bad bit-shifting and/or masking when translating the cylinder/head/sector information to and from the SCSI "logical block/sector" value.

More information HERE.


Known Problems

171 POST Errors

Symptom: A 00017100 (171) error occurs on POST (Power-On System Test). This problem may occur on 8595/9595 models xLx or xHx systems during the installation of a LANStreamer MC32, MC16 or EtherStreamer MC32 in slot 8.

Fix: Do not install any of the above referenced adapters in slot 8. Relocate the adapter to one of the other Micro Channel slots. No further engineering action is planned.

Diskette Data Loss (ECA100)

This problem may allow bad data to be read from or written to a device WITHOUT any error indications or other evidence of system failure. This affects any device attached to the floppy controller that is "seen" as a floppy drive, such as some tape backup devices.

Affected Systems: All systems using the 486SX/25 MHz complex FRU P/N 92F0079.

Notes: The following are no longer supported: 5.25" Diskette Adapter/A (6451007), 4869-001 360 KB and 4869-002 1.2 MB External Floppy. These adapters use a separate device driver (not BIOS), and DMA verify. Updating the system partition will NOT cure the problem.

16-bit Busmasters

When installed with the 486SX/25 Processor Upgrade Option, 16-bit bus masters (for example, PS/2 Micro Channel SCSI Adapter (#1005, 6451109)) that support 32-bits of addressing will cause system malfunction and/or potential loss of data when the user installs greater than 16 MB of system memory.

Content created and/or collected by:
Louis F. Ohland, Peter H. Wendt, David L. Beem, William R. Walsh, Tatsuo Sunagawa, Tomáš Slavotínek, Jim Shorney, Tim N. Clarke, Kevin Bowling, and many others.

Ardent Tool of Capitalism is maintained by Tomáš Slavotínek.
Last update: 03 Dec 2024 - Changelog | About | Legal & Contact