IBM SCSI w/Cache

Terminating the IBM SCSI-1 Adapters
OLD SCSI w/cache (Single oscillator)
NEWER SCSI w/cache (Dual oscillator)
NEWEST SCSI w/cache. (Triple oscillator)
   10 pin Artifact on Newest SCSI w/cache
Benefits of the new BIOS for the SCSI w/Cache
Swap BIOS for +1GB Support
BIOS Upgrade Needed for >1GB Support
Difference between -C188 and -C186 microcontroller
Specifications for SCSI w/cache (possibly OLD)
Results from upgrading to 2MB cache
   OLD SCSI w/Cache (Single and dual osc)
   NEW SCSI w/cache (Triple osc)
Installing 2MB
   Non-Compatible SIMM Results
   Converting Standard 30-Pin SIMMs
      8MB Cache Hack?
   Testing SIMMs for 2MB upgrade
OS/2 v.3 SCSI Controller switches
Autotermination Capability- Yellow vs. Orange Termpacks
   Terminator Resistance Values
   Termpack FRU
@8EFF.ADF - IBM PS/2 SCSI Adapter w/Cache 
C8EFF.ADF - Init file for _8EFF.ADF 
@8EFF.ADF - IBM PS/2 SCSI Adapter w/Cache
            (modified, no ADP, ROM area selectable / German comments)

Making Your Own IBM Cable
  The actual cable itself is a standard SCSI-1 cable. The unusual part is the 50 pin edgecard connector ( Dalco part 40720). Note that Pin 1 on the adapter is towards the mounting bracket . Remove the 50 pin dual row header (pry it open or cut it off) and crimp on the 50 pin edgecard connector. Watch pin 1!
   NOTE: On original IBM ribbon cables, the cable exits the edgecard connector TO THE LEFT as the connector is seated on the internal port. As installed, this means the cable goes straight up towards the top of the 95 case and away from other adapters. This makes it possible to route it away from them towards the front of the case, then down the site of the DASD structure, then to the drives.

   °   IBM Personal System/2 Micro Channel SCSI Adapter with Cache 

       The redesigned PS/2 Micro Channel SCSI Adapter Cache provides system configuration flexibility to support multiple internal and/or external SCSI devices such as high performance fixed disk drives, CD-ROM drives, tape drives, scanners, and printers. It has a burst transfer rate of 16.6MBps, adheres to the industry standard ANSI SCSI interface and is supported in either a 16- or 32-bit Micro Channel slot. 
    The IBM PS/2 Micro Channel SCSI Adapter with Cache is a 32-bit busmaster SCSI adapter containing a 512KB cache buffer that allows system memory to be totally dedicated to running the application rather than a portion being reserved for software caching. This SCSI adapter is recommended where improved data transfer rates and multiple SCSI devices are required and system memory is constrained.  However, 512KB cache is too small for the LAN Server application. 

OLD 6451018
F2  Bourns Transorb
U32 80C188 10Mhz-clocked busmaster cpu (20MHz divided by 2 for more stable waveform)
U44 Hitatchi HM62256LFP-12T
U47 Adapter BIOS 64F5984
U48 OKI 15F7917
U70 15F6903
U73 Adaptec AIC-6250EL Line Interface
U79 33F6715
U87-U88 30 pin SIMM
Y1 20.0 MHz Oscillator

Pin 2 of the internal port is on the component side. Pin 1 is on the circuitboard side.

  The "single OSC"-adapter is the "old" IBM SCSI Adapter with cache /A, which has no internal termination resistor and needs the large external Terminator block.P/N is 6451018 (FCC-ID:ANO6451018).. This adapter needs the update SCSI BIOS P/N 92F2244 / 92F2245 in order to handle drives >1GB.

Note: The Adaptec AIC-6250EL Line interface chip is just "glue logic". It is NOT the controller.
   While the C188 has (2) 8-bit parallel-ports it only has an 8-bit external data-path. 

ECA 032 Possible Replacement Required
   The old adapter w/cache can be identified by PN 64F1333 on module U47 (just below the MMKs). If you have this adapter, replace it.

Ed. I saw something about potential data loss concerning this ECA. Not sure- my U47 is the adapter BIOS.

U32 iN80C188
U47 Adapter BIOS, 33F5546
U48 15F7917
U70 15F6903
U73 AIC-6250EL Line Interface
U79 33F6715
U87-U88 30 pin SIMMs
Y1 20.0 MHz Oscillator
Y2 25 MHz Oscillator

FCC ID: ANO6451018 Mostly seen across the pond in Europe. This adapter needs the update SCSI BIOS P/N 92F2244 / 92F2245 in order to handle drives >1GB. The "dual OSC"-adapter is the "later" IBM SCSI Adapter with cache /A, which needs an external terminator.

Ed. Odd, I have a single osc 6451018, BUT it has the solder pads and the outline for Y2.

F2 Transorb
J300 C60 Port
T-Res SCSI terminator
U32 80C188-16
U33 84F8324 Line Interface
U44 SRM20256LM12
U47 BIOS 61G2976
U70 15F6903
U79 33F6715
U87, U88 30 Pin SIMMs
Y1 20.000 MHz
Y2 25.0000 MHz
Y3 32.0000 MHz
Functions of Late Adapter Oscillators
20MHz oscillator is for the SCSI-Line interface processor.
25MHz drives the cache controller. 
32MHz is for the 80C186-16 microcontroller, which supplies the busmaster functions. This card is significantly faster than the "old" adapter 

   The "latest" cached SCSI, FRU 85F0063 (FCC-ID is ANOSPRIME), has no Adaptec chip present and a larger yellow 20-pin DIL terminator chip close to the external port. This card has the strange 10-contact card edge connector close to the internal 50-pin card-edge. It has a third oscillator as well: a 32.0000MHz (Y3). 
   Based on Peter's experience, If you have one of the older IBM SCSI-Adapter or SCSI w/cache *cards* you can use the SCSI-Bios from the later-level SCSI Adapter with cache (92F2244 and 92F2245) to make the system capable of handling drives over 1GB IF the complex BIOS is of the correct type

New BIOS Improvements (From Tim Clarke)
> Oh, really? Please expound on the "extra features" of the new SCSI bios. In addition to handling drives >1GB as the IML drive, what else?

The upgrade provides the following additional capabilities:
  1. More efficient use of Adapter ROM memory - minimizes adapter configuration conflicts.
  2. BIOS support for fixed disks up to 3.94 gigabytes per device.
  3. Supports "Search IML" from any PUN (Physical Unit Number) or
     SCSI ID.  The system partition is no longer restricted to only SCSI
     ID=6.  Each fixed disk ID will be searched for a valid system  partition.
  4. Allows redundant system partitions (IML).
  5. Sharing of SCSI devices.  An external SCSI device, such as the 3511,
     may be shared between two system units. The menu item is "Target"

> So what good is the multiple, redundant partitions? Especially  with a T1-3 complex?
    If drive id. 6 fails 'recovery' can be automatic without the need to recover IML-capability by partially ripping the machine down to change SCSI Id.s.  You do, however, need to have planned for this and set up the appropriate 'IML/Reference Partition'(s) and 'Active Partition'(s) on the 'fallback' drives (Id. 5, 4, 3... etc.)

Microcode Differences
From Charles Lasitter
   With the triple oscillator (newest) versions of the Enhanced SCSI w/cache, I see some Microcode differences but don't see much in the way of any particular pattern for when one Microcode will be in use versus another.
   One adapter has 10G4890 dated 1991. Another has 61G2976 dated 1993.  What is the difference between these two microcodes? I invariably see the "1993" microcode with some unusual looking cache SIMMs, but changing them doesn't seem to make a difference. I see "1993" on some boards labeled as "CARD 1" and others" CARD 3".  There is nothing I can track there.
   I can't figure out why they would change it if they didn't make something better, but I don't know what that would be or how it would manifest itself. I know that the microcode from triple oscillator adapters can't be switched to the old unterminated adapters, but I'm uncertain about from one two another for the same "3-oscillator" family.

From Tony Ingenoso
   (guessing)  There may have been a change to deal with the low-power mode problems of certain microcode level Kazuza series drives...  Had to tweak the heads off the cylinder they were idled over to keep from losing data when the next write (after power restoration) happened to be on that
cylinder...  nasty problem...OS/2 had an ABIOS patch to deal with it on the pseudo-ESDI versions of the drive.

Roll Your Own
U47 Firmware - 27C512
U68 SCSI Bios - 27C256 (Even)
U69 SCSI Bios - 27C256 (Odd)

SCSI Device Ordering
   Older SCSI adapters supported only a single bus which provided both internal and external SCSI bus connectors for the single bus.  For these adapters, devices are logically ordered depending only on adapter slot  number and the SCSI IDs of the devices.

Maximum SCSI Devices Supported
  Narrow SCSI devices support 8 possible SCSI ID values; therefore, up to 7 narrow SCSI devices can be connected to the internal or external SCSI busses in any combination using the remaining ID values.  Wide and narrow devices may be mixed on the same internal/external bus by using the proper combinations of SCSI bus cables, terminators, and/or SCSI connector convertor adapters. 

 Is the SCSI w/cache SCSI-2?
The SCSI adapter on the 8595 is "SCSI-2 compliant" - means: it uses parts of the *command set* from SCSI-2 (command queueing, command tagging) but the *transfer speed* is SCSI-1 with 5MB/s. It is also a "Narrow"-SCSI controller - opposed to "Wide"-SCSI, which means it uses 8-bit devices and not 16-bit devices.
   However: it can handle even Ultra-SCSI drives, because SCSI is compatible in any directions - other than shit-IDE.
   And the controller is "Single Ended", which means that every signal has GND as return line. "Differential" means, that every signal has a return line with inversed polarity ... when -for example- +DATA0 sends a logical "1" on the active line the return line -DATA0 sends a logical "0". The transceiver circuit in the device checks the *difference* between the two signals (therefore "differential") - line disturbances afflict both lines at once and are eliminated therefore.
   Conclusion: the IBM SCSI adapter is 8-bit narrow, single-ended SCSI with SCSI-2 style command set and SCSI-1 speed of 5MB/s. It can handle 7 Fast- or Ultra-SCSI devices and can handle even 7 Wide-devices if there are converters used from Wide-to-narrow SCSI (68-to-50 pin) but these devices must use device numbers from 0 - 6. The SCSI-ID 7 is used for the controller itself. First device on IBM SCSI controllers is the one with the highest ID (6) opposed to Adaptec or Buslogic controllers.

Autotermination Capable Adapters or Planars
  If the card or planar has a red colored T-Res, it does not have the circuitry to support autotermination (you have to pull the termpack if you add an external device).  If the termpack is yellow, then the additional circuitry exists and you can add or remove external devices without having to pull the T-Res off the adapter.

> Peter, for the $64,000 Question. If you use a yellow termpack on a  adapter or board that had a red one initially, will that enable it  to autoterminate?

Definitely: No. The "Auto-Terminate" is a function that require a little more hardware: a switching transistor that disables the TermPwr wire from the T-RES and another transistor / IC function that senses the voltage on the TermPwr line and the voltage on the data lines to figure out whether the line is terminated or not.

Terminator Resistance Values
Fact sheet is HERE

Bourns 4120R-003, -221/331
Model (41 = Molded DIP), 20 Pins, R = Thick Film Low Profile, 003 = Dual Terminator
Resistance Code in the format R1 / R2,  First 2 digits are significant, Third digit represents the number of zeros to follow. 220 ohm, 330 ohm.

It looks like this..

Termpack FRU
From Tim Clarke
   AFAIK, there is only one FRU for the "internal" termination resistor pack for  the IBM SCSI w/cache (adapter FRU 85F0063) that has the appropriate 20-pin socket and  the IBM SCSI w/o cache (adapter FRU 85F0002).  The termination resistor pack (20-pin) is FRU 57F2870 and, provided you insert it with Pin 1 correctly oriented, should work O.K.

What about that Artifact?

From Ian Brown
   Interesting bit of history this. And no, you won't find it documented anywhere. IBM was one of the members of the organisation that originally set the standards for SCSI, and as usual they had their own ideas.
   One of these 'unique' ideas was the ability to be able to set the SCSI ID# remotely from the host adapter rather than having to set it on the device itself.
   The result of this was the 'legacy' small connector on some 'high end' SCSI hosts, and that awkward RS6000 socket as the external port, which became the IBM standard for a while.
   The idea was good, but be grateful it never took off, as I understand it required decoding hardware on the devices, which could have left us looking for special 'IBM SCSI' HDD's etc.
   It was never fully implemented as far as I know, SCAM overtook it.

  How did I find out? I went to a 'Big Blue' re-union recently, and met one of the original designers of the IBM SCSI host adapters. I'd been wondering about that extra connector for years, so I took the opportunity to ask him.

Specs on the SCSI w/cache
SCSI type SCSI-1 w/SCSI-2 commands)
SCSI bus path / speed 8 bit / 5 MB/sec
I/O bus path / speed 16 bit / 3 MB/sec
I/O features  N/A
RAID levels  None (use software)
Tagged Command Queuing  No
Old SCSI w/cache cpu 80C188 at 10MHz
New SCSI w/cache cpu 80C188 at 16MHz
Adapter Size  Type 3 (long)
SCSI Channels One (internal/external)
Connectors One internal; one external
Devices supported 7 devices per adapter
Cache std / max 512 KB / 2MB
Cache method  Two 30 pin SIMMs
Cache configurations  0, 512K, 2MB
Cache write policy Write-through

Pinouts for the SCSI /A and the SCSI w/Cache
  The edgecard for the SCSI Adapter /A has less contacts than the later SCSI w/Cache. Still wondering why... But you can see a tableized version (Thank you, thank you..) of Peter Wendt's observations. It is HERE.

Swapping BIOS chips for >1GB Support
From Peter Wendt
   Today - while juggeling around with parts & cards on a Model 90 - I decided by what reason to remove the SCSI-Bios from an old uncached SCSI and stuff that from the "later" cached on it. Earlier this year I'd tried to swap the entire Eprom-Sets among the old and the later cached - and it did not work. Today I left the busmaster microcode Eprom on the card and only swapped the SCSI-Bios. 

To my undescribeable surprise: it worked. To make sure that it is not a gimmick of that machines' Bios I tried it on a Mod. 70-A21, which has definitely no enhanced SCSI-Bios support in the planar microcode. I pulled the 2GB IBM 0664 harddisk from my WinNT Server and installed it in the Mod. 70. I have a heavily modified Mod. 70 - has a standard power-plug (as described on my page, folks !) and a Kingston 486DX-33 upgrade. I also installed the uncached SCSI with the old Eproms. No surprise: "No operating system" and the system halted. 

Then I used the 92F2244 and 92F2245 on that same adapter ... Voilà: "OS Loader V4.00 ..." and WinNT 4.0 Server started up ... ! (Ever seen that on a Mod. 70 ?) Tried the same procedure with the old IBM SCSI adapter with cache (the one *with* the AIC-6250EL Line Interface) - with the same results.  Old Bios-chips: No operating system - New chips: Win NT starts

Conclusion: If you have one of the older SCSI-Adapter *cards* you can use the SCSI-Bios from the later-level SCSI Adapter with cache to make the system capable to handle drives over 1GB.

I do *not* know if there is another limit after 4GB - but assume it is (I don't have drives over 2.2GB currently). This limitation will at least exist on the IML-machines, since the principle that starts up the IML cannot handle drives over 3.94GB (the mysterious IML-border) due to the technical method of putting the systempartition MBR at *the end* of the physical diskspace. The register width is obviously limited to any number of total data-blocks below 4GB. So that does not change at all.

Complex BIOS Upgrade Needed for >1GB IML Drive
From Tim Clarke
   After an extended E-mail exchange with Al Brandt, who couldn't get a machine to IML from a >1GB drive (SCSI ID. 6) attached to a SCSI-1 controller w/cache but with the notorious (now infamous?) 92F2244 and 92F2245 'Enhanced SCSI BIOS' ROMs using a Type-1 complex with either of the 'older' complex BIOS ROMs (i.e. 84F9154 for SOD Type-1 and 91F9812 for non-SOD Type-1), I decided to run my own tests. 
   'Enhanced IML' in the complex upgrades is the *only* way to be able to IML from a drive >1GB (and from a drive Id. ¬= 6).
   At this time I must assume that the 92F2244+5 ROMs provide support for drives <= 4GB, but the 'old' complex ROMs' IML support somehow does not make (proper) use of it, 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.

From another thread
    BTW, space is usually allocated as a whole no. of cylinders, so the 'old' 3MB "System/Reference Partition" will grow to ~12MB on a >1GB drive.Albeit that only the 1st 3MB of that space is used/needed.

Some Other Thoughts
From Charles Lasitter
   I've had some VERY entertaining results in my installations, depending on what other drives were present, and which version of the processor BIOS was used.
   With the 52G9509 in place, ID6 in bottom bay (Mod 95) at end of cable, ID5 in bay above on next spot on cable, I couldn't get the IML to go to ID6 to save my ass.  It made a beeline for ID5 every time.  Put in the old BIOS, and it goes straight for ID6.
   I think there are a LOT of quirks like this (and yours) to be mapped out, and that seemingly innocuous settings changes in the ABIOS make differences you'd never guess sometimes.
   I also suspect that the Mod 90 is it's very own distinct bird with it's own eccentricities in this and related matters.

2MB SCSI w/cache

Can I increase the cache to 2MB?
One module that can be used to expand the IBM SCSI Adapter with cache /A from 512K cache to 2MB cache is the IBM P/N 30F5360 / FRU 74X8637, which is a 1MB x 9 bit (Parity) 100ns module *with* IBM-specific presence detection.

The above-mentioned IBM modules are in the Mod. 30-286 (8530-Hxx or -Bxx) with 80286-cpu - if anywhere at all.

MY 1MB SIMMs are marked MSC2314-12YS9A 183004 68X5721.
It has 9 OKI M511000A-1AJ chips, mnf. 89335521

From Jerry Dumer
I have 3 boards with Toshiba THM91010AS-10 simms that do the job fine. I found these on a memory card in a Model 60. They are gold pins. I imagine Toshiba made them for the 30-286s. They do work.

Kingston modules labeled KTM-1000/M30 (From Peter)

Trying Non-Compatible SIMMs
   If you try some likely looking 30 pin SIMMs and they don't have the correct CAS/RAS wiring, the system will disable them it and you will see SCSI Adapter w/Cache with 0KB. Don't freak out. Replace the original SIMMs and run Advanced Diags and test the SCSI Adapter to restore your 512K cache. Been there, done that.

Can Industry Standard Memory Be Converted?
   Yes. Alfred Arnold has finally figured it out. Dr. Jim Shorney confirms this triumph. The biggest hurdle was the IBM modules have a different RAS/CAS scheme in addition to the different pin-out. As time goes on, exceptions MAY be found. News as it happens

Alfred's 30-pin SIMM Hack (To his site)
   For those that have the burden of college education, I made some assumptions. First, the SIMM pins will always  be performing the same function (like parity Data Out, parity CAS). Second, those simm pins can be used to positively identify the parity chip's DO and CAS pins.
   This makes manufacturer's data sheets uneccessary. Assuming they even have them.

break the connection between SIMM pin 26 and the parity chip's DataOut pin; This inferrs that SIMM pin 26 goes to Data Out- 

connect the parity chip's DataOut pin to SIMM pin 29;
Tie the parity chip's lead that USED to go to SIMM pin 26 to SIMM pin 29

break the connection between SIMM pin 28 and the parity chip's CAS pin; 
Parity chip CAS is attatched to SIMM pin 28

connect the parity chip's CAS pin to SIMM pin 2; 
Attatch lead that used to go to SIMM pin 28 to SIMM pin pin 2

connect both SIMM pins 26 and 24 to SIMM pin 22 (GND).
SIMM 26 and 24 tied off to SIMM 22

A picture of Dr. Jim's 2MB Hack

8MB Hack?
   Also: No. It is a little miracle that the 1MB modules work for the 2MB total cache, but the 4 Meggers are -again- different to the 1MB ... apart from the presence detection and the RAS/CAS scheme. There are 1MB modules that behave the same way than the 256K modules - but no 4MBs. 
   The 4MB you find are either Industrial Parity (9 bit, but w.o. presence detect) or non-Parity Industrial modules (8-bit). IBM itself never introduced 4MB modules of that scheme for their own machines. The PS/VP Series 1 used 1 and 4MB 30-pin modules, but these were "industrial standard" modules. 
   The PS/2 256K, 512K and 1MB modules are very unique. They depend -technically- on a very old (1985) Hitachi patent to which IBM only added the presence detection. From the principle the "Single RAS" scheme of these modules is similar to the Hitachi HB61009BR-15, 150ns module - a very antique construct, which were superseeded by the much more common HB41256-style design, which uses separate RAS/CAS timing for interlaced access to the cell-array, which then allowes much faster RAM access (typically 100 - 60ns).

Testing SIMMs for use in SCSI Adapter w/cache
  The secret is out... What to look for if $10 for two SIMMs is too much for you...

Results of upgrading
From an article by Peter Wendt....
The IBM Adapters with and without cache /A are in fact SCSI-1 adapters with 5MB/s interface speed (!) to the device. This however does not tell how high the effective data-throughput might be.

This depends on more factors than just the interface-speed:
- speed of the drive mechanism (track-to-track / average seek)
- adapter-cache (size, speed and degree of optimizing)
- device-cache (size / usage)
- cable and degree of disturbances through line noise / spikes etc.
- termination (reflection of data packets)
- adapter- and device command set (autonomy of device / command queing)
- the adapters chipset

Okay - for the curiousity: I have tested the 2MB-cache in my 8595-AK9 (32MB RAM / Kingston TurboChip / modified BIOS) with 2 different IBM adapters and a Quantum Fireball TM2110S (2.1GB Ultra-SCSI). This drive is known as quite fast and should not be the bottle-neck. I used the IBM SCSI adapter with cache /A in the version *without* the internal termination resistor and the Adaptec AIC Line interface chip (IBM old) and I used the IBM SCSI Adapter with cache /A *with* the internal yellow termination resistor and the IBM SCSI Interface (IBM new).

Here are some test-samples (results are temporary examples)
Results for old SCSI w/cache
IBM (old) 512KB cache 2MB cache Change (%)
Minimum (kB/s 28.8 28.9 + 0.3
Weighted Average (KB/s) 796.4 874.8 + 9.8
Maximum (KB/s) 1608.4 2060.5 +28.1

Results for new SCSI w/cache
IBM (new) 512KB cache 2MB cache Change (%)
Minimum (kB/s) 46.9 46.5 - 0.8
Weighted Average (KB/s) 1188.3 1285.1 + 8.1
Maximum (KB/s) 2122.1 2579.9 +22.9

Results obtained by reading and writing blocks from 512 bytes to 63.5KB in linear and random methods.
Read Linear with large data-blocks shows a 20% gain, due to the 2MB cache.
Write Linear values are only 2 - 5% smaller in either case.
Write Random values show the influence of the drive mechanism.

What does this mean in practise ?
1.) Only way to get minimum values up is to use a harddisk with a faster drive mechanism..
2.) Combination drive - adapter doesn't reach the data-throughput for the SCSI-1 interface speed.
3.) Any GUI operating system that uses a temporary swap file operating with large blocksizes will make use of the enlarged SCSI-Adapter cache during the swapping process.
4.) Effects are noticeable in Win95 applications and -probably- much more noticeable on OS/2 and NT, which use the harddisk-subsystem more intensely than DOS/Windows or Win95. (That's a prediction from my experiences with these OS)
5.) If you cannot -by what reason- use a F/W-adapter / F/W-drive combination (or don't want to) the enlarged cache to 2MB is one simple method to get out around 10 - 20% better performance of an unmodified system running under i.e. Win95 (I'd kept this careful - you cannot await wonders if you have a slow harddrive)

 The testing will continue as soon as possible. Like very often my job came in the way and I needed the 95 (which is usually my network-server) for the daily purposes again. But this first testing suite shows that the enlarged cache *has* in fact a good impact on the over-all performance of the system. I'll try to fix up the results and try bringing them here - if you like "data cemetries"... the tables are *very* complex and have hundreds of numbers, which must be set in proper relation to each others. And must be interpreted the correct way.

Controllers Main Page

9595 Main Page