While putting together the 486 DX4/DX5 build from page 4, i encountered some problems around SCSI and WB1.
This prompted me to look for alternatives such as using VLB IDE controllers instead.
Until then i never spent the time to find-out who is who in this department, but my desire for fully maxed-out system required some exploration.
Sharing here the collected data with the hope that it will be useful to others.

HARDWARE

(Adaptec AIC-6360L for SCSI, Adaptec AIC-25VL01Q for IDE) Hornet Technology VL-300, details here.
(Appian ADI2) BCM GP-4530
(ALi M5213, relabeled Appian ADI2) PALM VL-BUS revision B.3
(ATronics IDEC-2020L) DTC 2177A, 16Mb RAM
(ATronics IDE-2015PL) Boca Research IDEVL2
(ATronics IDE-2015PL) unknown brand/model
(BusLogic BA69C10, Harris CS80C286-20) Buslogic BT-410A / (Tekram TRM680C, Harris CS80C286-20) Tekram DC-680C
(CMD PCI0640B) CMD Technology Inc. unknown model
(Compass Lab CL3202, relabeled Chips 82C712 ?) unknown brand/model
(Dataworld ADD2) Unknown VLSIO2C3
(DTC DTC2278E) Data Technology Corporation DTC2278E
(DTC DTC-805 B) Data Technology Corporation DTC2278 VL
(GoldStar GM82C712, relabeled Chips 82C712) unknown brand E852
(Holtek HT6560B) Holtek VIO3B-V1.1
(LRC-03) Lion 3+
(NCR 53CF9402) UltraStor Ultra 34FB
(OPTi 82C611) OPTi VIC
(OPTi 92C108, Intel N80C186XL-12, Qlogic FAS216) DTC 3274
(PiC 12885A-125) Astral Corp VLMIO v1.6
(Promise PDC20630) Promise EIDE2300Plus
(Promise PDC20630) SIIG SC-VKA422
(Promise PDC20630) Promise P2630VL-1
(Promise PDC20330A, PDC20430, NCR 53CF94-2) Promise DC440, details here.
(Promise PDC20230C) Promise DC-4000
(Promise PDC20230C) FIC Combo-542VP
(Promise PDC20130A, PDC20230B) Promise DV4030VL-1
(QDI Vision QD6580) QDI QD6580UMC V1.0
(QDI Vision QD6500) QDI QD6500A v2.0
(QLogic FAS216) Buslogic BT-445S
(SiS 83C611) Tyan S1342/1344-005
(SiS 83C611) Tyan VL400GW-V1
(SiS 83C611) Tans SVIDE-V02
(ST300ALI, AMD 80C186-16) DTC 2277VL
(ST300ALI, AMD 80C186-12) Tekram DC-680T
(Tans TS8310) Tans TS-8310VLT VB.1
(UMC UM86C473F) TK TK-85C418GVIO 4J D08
(UMC UM82C872F) unknown brand/model
(UMC UM82C871F) unknown brand/model
(VIDE-1) Astral Corp VLMIO V1.4
(Winbond W83759F) Micro Equipment Corp UN-1082 VL

NOTES

All controllers run at their fastest speed settings (jumpers and driver arguments). Missing bar in the charts below indicates unstable system - usually permanent data corruption or inconsistent reads.
With the next exceptions:
- DC-4000 can only operate in "normal" speed (other options are "fast" and "turbo"). Tried on few different motherboards - no bueno, so using it like that.
- Couldn't find drivers for TK-86C418VIO, so the captured performance is most likely not indicative to what the hardware can actually do.

UMC UM82C872F's driver is really interesting, it can spawn a sub-menu during config.sys execution. Using that menu user can interactively adjust its settings.

UM82C871F seems like a bad adapter. Does not like CF cards, mechanical HDDs only. It requires you to format the HDD through it, otherwise cannot boot and/or if it sees a partition most of the files/directories appear corrupted. Maybe there is a logical explanation behind this behavior, but didn't bother to check, maybe later. For now i won't include it in the test charts, because i cannot test it against the same CF.

DTC2278E appears to be the latest model of series of IDE/EIDE controllers from Data Technology Corporation. It is an EIDE model with full support for PIO 4 mode.

Couldn't find IDE driver for DC440.

Couldn't find IDE driver for VL-300.

When not using the software driver for the SIS controller, both buffered and linear read speeds are at ~4100 Kb/s. When use the driver, the buffered read goes through the roof, but the liner speed increased only by 700Kb/s = ~4800 Kb/s.

Compass Lab CL3202 and GoldStar GM82C712 seem to be Chips 82C711/712 in disguise. That is IDE PIO-3 silicon.
711 is jumper configured, 712 is software driver configured, or something like that. Both paths converge on the same capabilities and performance.
At the same time the goldstar drivers don’t work on the Compass Lab controller, so the above presumption may be incorrect.

DTC 2177A takes long time to complete POST and BOOT to DOS.
Manual incorrectly states to set drive type to 1. It must be CUSTOM with the actual drive settings. Ideally use NORMAL mode, but not LBA.
No drivers are required. On-board BIOS takers care of business.

It does not look like the driver for (Dataworld ADD2) VLSIO2C3 is actually doing anything.
With, or without it, performance is the same. Further investigation will be needed.

There are 2-3 driver versions for the ATronics IDE-2015PL based controllers.
The late drivers offer control parameters, but seem to be problematic in general. Performance is really bad. Way below 1Mb/s. Can be related to DMA access. Not sure.
The early driver does not offer control flags, but boosts performance about twice.

Using the /H flag in the the driver of CMD Technology Inc. (CMD PCI0640B) causes write issues with CF cards. The presence/absence of that flag does not seem to affect performance.
When the /A flag is set to 20 or 33 (reduced wait states) and FSB is 40MHz or higher - performance is impressive (fastest controller to date), but similar to when /H is used, write issues with CF cards occur.

Lion 3+'s DOS driver makes the system unstable with CF cards.
From what i can see it is the software counterpart of the speed jumpers, so no big story here. Ran the tests without it.
Manual and drivers:

OPTi VIC - the 32-bit access driver (OPTIVIC.SYS) causes problems with CF cards - FAT corruption.
No permanent damage, but directory and file structure gets mangled for the session upon driver load.
The SETUPVIC.EXE and its command arguments does not seem to make any difference, except the 25/33/40MHz one.

(SiS 83C611) Tyan VL400GW-V1
Up to PIO-3.
Wait states can be controlled through jumpers or software driver.
Driver has setting for multi-word I/O that does nothing with CF cards. Didn't test with period correct mechanical HDDs.

(Promise PDC20630) SIIG SC-VKA422
A Promise EIDE2300Plus in disguise, with customized BIOS from SIIG.
That custom BIOS makes the controller initialization process (during POST) significantly longer than the one of EIDE2300Plus.
It is so long that one can think the computer hung. The EIDE2300Plus drivers work directly with that card, obviously.

(QLogic FAS216) Buslogic BT-445S
Unstable with wait states set for 20 and 25MHz base frequency.
Wait states set for 33MHz is the minimum for stable system.
Switching between 5 and 10 Mb/s option does not change performance.
Using BTDOSM.SYS driver with /N:02 and /F:01 boosts transfer rates by about 1Mb/s.

(ALi M5213, relabeled Appian ADI2) PALM VL-BUS revision B.3
I was not sure what the ALi M5213 chip is, or what drivers were for it, but found images of the same VL-BUS card with Appian ADI2 IC on it.
The Adaptec/Appian FLEXIDRV.SYS works with it which made it clear what's under the label. Didn't list the results explicitly in the charts below top reduce clutter since they match Appian ADI2.

Tekram DC-680T and DTC 2277VL are basically the same controller with different BIOS versions.
Also, DTC put 16MHz AMD 80C186 processor while Tekram used 12MHz one.
The DTC BIOS is clearly better. Supports LBA mode, also much faster at POST/BOOT.
The Tekram BIOS is more prickly and can lead to instabilities with highly optimized systems.

Promise DV4030VL-1
Default BIOS does not support LBA, so if the motherboard's BIOS has the attached local storage devices specified through LBA, the controller won't be able to see them.
Later BIOSes (available on the forums here) solve the issue.

DTC 3274
No UI, switches and jumpers only.
Upgraded the on-board cache from 16Kb to 64Kb. This didn't have any effect on performance.

THE CULPRITS

(Adaptec AIC-4310P) Adaptec AHA-2840-A

(Adaptec AIC-6360L for SCSI, Adaptec AIC-25VL01Q for IDE) Hornet Technology VL-300

(ALi M5213, relabeled Appian ADI2) PALM VL-BUS revision B.3

(Appian ADI2) BCM GP-4530

(ATronics IDEC-2020L) DTC 2177A

(ATronics IDE-2015PL) Boca Research IDEVL2

(ATronics IDE-2015PL) unknown brand/model

(BusLogic BA69C10, Harris CS80C286-20) Buslogic BT-410A / (Tekram TRM680C, Harris CS80C286-20) Tekram DC-680C

(CMD PCI0640B) CMD Technology Inc. unknown model

(Compass Lab CL3202) unknown brand/model

(Dataworld ADD2) unknown VLSIO2C3

(DTC DTC2278E) Data Technology Corporation DTC2278E

(DTC DTC-805 B) Data Technology Corporation DTC2278 VL

(GoldStar GM82C712, relabeled Chips 82C712 ?) unknown brand E852

(Holtek HT6560B) Holtek VIO3B-V1.1

(LRC-03) Lion 3+

(NCR 53CF94-2) UltraStor Ultra 34FV


printed manual

(PiC 12885A-125) Astral Corp VLMIO v1.6

(Promise PDC20630) Promise EIDE2300Plus

(Promise PDC20630) SIIG SC-VKA422

(Promise PDC20630) Promise P2630VL-1

(Promise PDC20330A, PDC20430, NCR 53CF94-2) Promise DC440


printed manual

(Promise PDC20230C) Promise DC-4000

(Promise PDC20230C) FIC Combo-542VP

(Promise PDC20130A, PDC20230B) Promise DV4030VL-1

(QDI QD6580) QDI QD6580UMC V1.0

(QDI Vision QD6500) QDI QD6500A v2.0

(QLogic FAS216) Buslogic BT-445S

(OPTi 82C611) OPTi VIC

(OPTi 92C108, Intel N80C186XL-12, Qlogic FAS216) DTC 3274

(SiS 83C611) Tyan S1342/1344-005

(SiS 83C611) VL-IDE VL400GW-V1

(SiS 83C611) Tans SVIDE-V02

(ST300ALI, AMD 80C186-16) DTC 2277VL
New in the box.

(ST300ALI, AMD 80C186-12) Tekram DC-680TS

(Tans TS8310) Tans TS-8310VLT VB.1

(UMC UM86C473F) TK TK-85C418GVIO 4J D08

(UMC UM82C872F) unknown brand/model

(UMC UM82C872F) unknown brand/model

(VIDE-1) Astral Corp VLMIO V1.4

(Winbond W83759F) Micro Equipment Corp UN-1082 VL
=

BENCHMARKS

Tested on Asus VLI-486SV2GX4 rev.2.0 motherboard with patched BIOS for WB cache, 1024Kb SRAM, 64Gb RAM, AMD 486DX5 ADZ CPU running at 160MHz (4x40).
Added Adaptec AHA-2840-A VLB for a reference, marked in white outline.

Used CF card several generations faster than any of the tested adapters, to make sure they don't get bottlenecked on that side.

SpeedSys ranking is based on the sum of buffered + linear read speeds.
(thinking to revisit that for the note below)

Gray bars denote controllers with large discrepancy between buffered and linear read speeds.
In SpeedSys, for caching controllers, the long bars (buffered reads) indicate cached transfer speeds, short bars (linear reads) indicate the actual non-cached transfer speeds when the storage device is accessed directly.
Simply put - the short bars represent the overall performance of the adapters.

Private conversation with jakethompson1 prompted me to test with mechanical HDD as well.
I picked Maxtor 90650U2 (MA540PR0), 5400 rpm, 6.4Gb, Ultra-ATA/66, 2Mb cache.
The UMC and Promise 20630 drivers didn't like it for some reason. Clunky ...

CONCLUSION
First of all there is some homework left to do - run more in-depth perf tests that can reveal additional details, but that's for later.
For now - take a closer look at the Holtek, Appian ADI2 and DTC controllers.

Never heard of Holtek before, but it totally kicks ass.
It shows write access issues with SanDisk Ultra CF cards, but works well with Transcend and others. Still, something to keep in mind.

@Warlord provided a great tip that Appian ADI2 chip is powering the Adaptec AVA-2625VL EIDE controller. He also provided drivers. Thanks.
The controller is not completely stable with mechanical HDDs. I am testing on a highly optimized system (which is a part of the test itself) and some adapters struggle with that.
But since i am on the CF cards bandwagon - that's not much of a factor.

The best performers are DTC2278E, Holtek HT6560B, Promise EIDE2300 Plus.
Holtek is not completely stable with CF cards - sometimes misses write access and asks user for "retry", then completes the operation just fine.
DTC and Promise are completely stable and work equally well with CF cards and mechanical HDD.
DTC hangs with POD83/100 + L1 cache in WB mode. Same happens with the SiS controllers, btw.
So at the end of the day the true hero is Promise EIDE2300 Plus.

Despite the seemingly faster sequential speeds when using mechanical HDDs, in reality they are much slower than CompactFlash cards because of slow seek and random access times.

Updated the perf charts on pages 4 and 5 with with new VLB VGA and IDE controllers.
New pictures on page 1.

----------------------------

HAM/12-W2 (KT216WB) - an unimpressive 286 motherboard.

Cannot work reliably over 16MHz.
Can see maximum 4Mb of RAM.
Couldn't find documentation for it other than a page on stason about similar mobo, but different enough.
There are couple of jumpers - tried different configurations - nothing seems to change.
The board refuses to recognized CF cards, so had to strap a mechanical HDD.

The Wolf3D test quits before it finishes.

https://petershipkov.com/temp/retro_pc_images … motherboard.jpg

This motherboard is known as KT216WB (stands for KT2, 16mhz, WinBond?)

They really don't work above 16MHz, is overclocking. The initial normal mode is 12 MHz. They were sold at the beginning as 12 MHz, then the manufacturer clocked up to 16 MHz.

From the Russian site, Jumpers:

JP2: 1-2 к pin33 Jetkey BIOS to +5V через 1кОм; 2-3 к pin33 Jetkey BIOS to GND; (возможно выбор режима работы клавиатуры) === JP […]
Show full quote

JP2:
1-2 к pin33 Jetkey BIOS to +5V через 1кОм;
2-3 к pin33 Jetkey BIOS to GND;
(возможно выбор режима работы клавиатуры)
===
JP5: (зачем надо непонятно, ни на что не влияет)
pin1 к pin23 Jetkey BIOS;
pin2 к GND;
pin3 к pin24 Jetkey BIOS;
===
JP9 - parrity error check On/Off (использовать или не использовать памяти четности 4xDIP16)
===
JP10: Power Good
1-2 Detect from power supply;
2-3 Detect from board;
===
JP11: (получается актуальная настройка сигнала power good для древних блоков питания)
pin1 - power good;
pin2 - через 470 Ом к pin3 JP10;
===
JP12 - Read RAM wait state 0/1
JP13 - FPU clock Fast/Slow (включает или выключает делителя от основного клокера для сопра)
===
Память:
JPA1; JPB1; JPC1 - Memory control:
A1:
1-2 Минус 128 КБ на нужды системы, наверное или EMS Memory, или shadow RAM, или активировать UMB;
2-3 - OFF;

B1:
1-2 Включение второго банка памяти;
2-3 - OFF;

C1:
1-2 - Объём каждой микросхемы памяти 256х1 кБит (или кратно этому числу если 4-х битная, определяет тип матрицы каждой микросхемы);
2-3 - Объём каждой микросхемы памяти 1024х1 кБит (или кратно этому числу если 4-х битная);

В итоге если принять что соединение пинов 1-2 - это "0", а 2-3 - это "1" получается следующее:
A1 B1 C1
0 0 0 - 512кБ;
1 0 0 - не поддерживается;
0 1 0 - 1024кБ;
1 1 0 - 896кБ;
0 0 1 - 2048кБ;
1 0 1 - 1920кБ;
0 1 1 - 4096кБ;
1 1 1 - 3968кБ.
===
JPA/B/C - Программный лимит подсчета памяти до: 512кб, 640, 896, 1024кб, на большее не проверял.
Дополнять SIPP/SIMM бесполезно, пока не снимите все DIP-памяти, т.к. разводка абс.параллельны.
Примеры (вид сверху):
1 Mb RAM = JPA 2-3, JPB 1-2, JPC 2-3
2 Mb RAM = JPA No, JPB 2-3, JPC No link: https://pp.userapi.com/c841531/v841531640/2ed … MRVXPPYI3e4.jpg
4 Mb RAM = JPA 2-3, JPB 1-2, JPC 1-2 link: https://www.phantom.sannata.org/download/file.php?id=1916

Заменен резистор R26 на 5кОм, вместо штатных 200-400 ом. Это улучшает совместимость со всеми клавиатурами (согласно ГОСТу-стандарта согласования входа-выхода клавиатурного разьема DIN-5). Резистор R26 находится возле разьема сопроцессора.

Good info.
Will give J12/13 a spin over the weekend.

What is the web-site you got this info from ?
It certainly didn't show in my search results.

Thanks.

One of the popular Russian-language forums in the CIS is https://www.phantom.sannata.org/search.php?se … d=active_topics

In any case, we only know the motherboard KT216WB, not otherwise. I think that Google Search will give much more info for KT216WB 😀

Here you can see the original board with 12 MHz, as it should be. No factory overclocking.

@ekb
Thanks for the link. Interesting. Do you know if sellers ship internationally ?

------------------------------

Biostar MB-1333C-CH/1340C-CH

I usually avoid wide format early 386 motherboards. They are too slow and "rigid" - pain in the bottom to deal with.
This one is no different. Been bugging me for long time.
I saw some topics on the forums going through that stuff, but they appear inconclusive to me.
Decided to check if somebody around figured it out already.

The board supports 2 SRAM configurations - 64Kb and 256Kb. Respectively 16kx4 and 64kx4 chip types, which are not common, or at least i don't posses many. I have 64Kb installed on it atm.
BIOS is littered with timing options and all of them except two can be set to their lowest value. Great stuff !
No, not really. Cache is not recognized no matter what.
There is a chance some of the cache chips are faulty, but the mobo does not report "bad cache" which lowers the chance of it, but does not eliminate it.
Also, the installed cache limits the frequency to 33MHz max. Anything above that with the cache enabled from the BIOS and POST never completes.
So, at least to some extent SRAM is functioning.
I am hesitant to spend coinz on this kind of cache chips - they are way too expensive for what they are and of little use outside of very few boards that use them.

Any 3 chip SIMMs are no-go with this board. It works with parity 9 chip SIMMs only.

It seems that there are at least 2-3 other motherboards which are basically identical - Peak and some others.
Anyone got one of them working with functioning cache for real ?

Updated the first post in this page with few more VLB IDE controllers.
Tested all of them with mechanical HDD in addition to the CF card stuff from before.
Some interesting results.

------------------------------------------------------------------

Spent some time today with Alaris Leopard revision C based on (OPTi 82C295) - an IBM BL2 processor ready motherboard.

There is plenty of information about IBM BL2/BL3 CPUs, so wont repeat any of it.
The whole thing is unusual - a fascinating hodge-podge of different technologies fused together.

The motherboard is of high quality and well preserved - like new.
It can handle up to 16Mb of RAM and 128Kb SRAM. The type of CPU this board is built around can cache 16Mb only anyway.
Installed 4x4Mb 60ns SIMMs and 2 12ns cache chips to max it out.
BIOS is full of options. All set to their optimal values for best performance. The board takes it with ease.
CPU is running at 80MHz. Heat sink is too small to cool it down once under load, so proper ventilation is required, otherwise the system can hang.
Used Ark1000VL VLB video card - seems to be the fastest thing for DOS graphics and will help the system show the best it can do.
Many VLB EIDE/SCSI controllers didn’t work properly, the few that did (such as DTC 2278E) required lowered settings for complete stability.

Have a second board of the same type but with slightly different aesthetics - black resistor arrays instead of red.

Small detour:
That second board came with IIT FPU. This allowed me to compare it against the ULSI above.
Turns-out ULSI is same or faster than IIT depends on which test we look at.

Ok, back to the main subject.

The usual set of stats. Looking pretty good.

benchmark results

DOS interactive and Windows accelerated graphics look pretty good.
Things are even better in the offline compute and FPU heavy tasks.

Tight package overall, not very picky.
Limited upgrade options.
Overall - pretty good hardware.
Best part - these boards still come for sale from time to time at reasonable(ish) prices.

---

Back in the day i had no idea about any of these BL CPUs and Alaris boards.
The experience today got me thinking - who was the target audience for this kind hardware ?
Some internets suggest it was the early 90ies enthusiasts ...
Kind of unlikely in my opinion. Performance is lacking compared to 486 DX2 systems.
Maybe Alaris Cougar + BL3 at 100MHz for enthusiasts ?
Maybe.
Wonder what was Alaris's business model and what market niche they aimed at.
But still - for what it is - a really fascinating piece of retro technology.

pshipkov wrote on 2020-08-16, 23:48:

@ekb Thanks for the link. Interesting. Do you know if sellers ship internationally ? […]
Show full quote

@ekb
Thanks for the link. Interesting. Do you know if sellers ship internationally ?

------------------------------

Biostar MB-1333C-CH/1340C-CH

I usually avoid wide format early 386 motherboards. They are too slow and "rigid" - pain in the bottom to deal with.
This one is no different. Been bugging me for long time.
I saw some topics on the forums going through that stuff, but they appear inconclusive to me.
Decided to check if somebody around figured it out already.

The board supports 2 SRAM configurations - 64Kb and 256Kb. Respectively 16kx4 and 64kx4 chip types, which are not common, or at least i don't posses many. I have 64Kb installed on it atm.
BIOS is littered with timing options and all of them except two can be set to their lowest value. Great stuff !
No, not really. Cache is not recognized no matter what.
There is a chance some of the cache chips are faulty, but the mobo does not report "bad cache" which lowers the chance of it, but does not eliminate it.
Also, the installed cache limits the frequency to 33MHz max. Anything above that with the cache enabled from the BIOS and POST never completes.
So, at least to some extent SRAM is functioning.
I am hesitant to spend coinz on this kind of cache chips - they are way too expensive for what they are and of little use outside of very few boards that use them.

Any 3 chip SIMMs are no-go with this board. It works with parity 9 chip SIMMs only.

It seems that there are at least 2-3 other motherboards which are basically identical - Peak and some others.
Anyone got one of them working with functioning cache for real ?

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

Even with 12 ns 64kx4 SRAM, I wasn't able to get 256 KB working properly on this motherboard.

Thanks Feipoa, good to know.

------------------------------------------------------------------------

Added few more VLB IDE/EIDE controllers to the first post on this page for completeness.

------------------------------------------------------------------------

Conversation with Annonymous Coward prompted me to try TI 486SXL2-50 CPU on the 386 Symphony system. Credit goes to Feipoa for connecting me to somebody who was selling the processor.
Results are pretty great. Updated the related post Re: 3 (+3 more) retro battle stations accordingly (towards the end, under the dashed line).
The system targets year 1993, but consider moving it to 1994 to take advantage of the SXL2 CPU and CL-5434 ISA video card, for top performance ...

------------------------------------------------------------------------

Examined recently a MG branded 286 motherboard.
I don't know anything about MG, other than it is a Japanese company.
Also, couldn't find any info online about it (the mobo), so no idea about exact model.

The board is a looker and really well preserved - crisp.
It came with SIPP memory slots. Did the usual trick and inserted SIMM ones into them. They fit really tight. In fact too tight, but that's a good thing.
There are no jumpers other than 2 for memory parity check and a switch box that controls the few RAM configurations.
Installed crystall oscillator socket.
As i said - couldn't find documentation online, so had to experiment a bit to find the right set of switch states for 4Mb of RAM.
The system cannot handle more than 20 MHz, but ticks pretty well within this range.

The standard set of stats.
Screen capturing didn't work (red flag), so snapped pictures of the screen.

benchmark results

All in all, not a bad piece of hardware, but there is better stuff out there.

You should be able to get 55 MHz out of the sxl2 . Fpu fully stable at this speed is tough though. Even an IIT x2 50 when pressed will fail

Yes, sxl2 can tick with 55mhz, but without FPU and RAM reduced to 4mb.
These limitations combined with the overall instability of the system exceed my tolerance threshold 😁
Still, it is interesting to see how far we can roll with 386 class hardware.

I've run the SXL2-50 at 55 Mhz on quite a few 386 motherboards with 32 MB of RAM without issue. I have the ULSI DX2-66 FPU, so the FPU limitation isn't so much of an issue.

It's important that you set the registers up properly for L1 and L2 cache when working with an SXL.

I didnt know that there is 66mhz 387 fpu 😀
I used the next flags: cyrix.exe -e -r -i1 -i2
But i know that i didn't spend enough time on it, so there is ground to be covered for sure.
Do you have a good formula about the flags ?
It felt that it is a case by case thing.
Thanks.

Depends on if you are using Windows or only DOS. Also depends on if you are using the Adaptec 154X or the 152X. I suggest reading the first section of this: Register settings for various CPUs

Thanks for the link. I remember reading it before.

Spent some time this morning to try the SXL CPU on few 386 boards according to your notes in that link - DTK (Symphony), ASUS (SIS rabbit), FX-3000 (UMC) boards.
Things didn't work out with the ASUS board.
The Symphony board takes the processor to 55MHz, but unstable. When Doom finishes (once out of # runs), i get 20.1 fps, which is pretty good, but has no purpose outside of this fragile overclocking experience. 😁

Anonymous Coward suggested OPTi SX chipset, so i got one on the way, will see how it goes.

--------------------------------------------------------

Checked some late (1994) ISA video cards to see if VGA graphics have any impact on the stability/performance of the above experiment.

HARDWARE

Diamond Speedstar Pro rev. A2 (Cirrus Logic GD5429)
Diamond Speedstar 64 ISA rev. A3 (Cirrus Logic GD5434)
STB Nitro ISA (Cirrus Logic GD5434)
Paradise Accelerator 24 (WD90C33-ZZ)
ET4000/W32i (late model) - this is my default VGA ISA adapter for 386 activities

NOTES

The performance difference between the 4 cards on a 386 hardware ranging from 40MHz to 55MHz was within the rounding error.
It was obvious that my usual tests are not VGA performance bound with these video adapters, so decided to go generation or few above and test on an ASUS VLI-486SV2GX4 board with DX5 CPU running at 160MHz, so perf gets bottlenecked in ISA/VGA.

Along the way i noticed few other things worth mentioning:
1. The Diamond 5434 card had troubles dealing with 0-wait-state with some of the 386 motherboards.
2. WD90C33-ZZ crashes in Wolf3D if the system timings are too tight, as is the case with the Asus VLI rig. I noticed the same with the VLB adapter - works well most of the time, but tends to get prickly if pushed. In the other hand picture in graphics mode is great, but initially got some skewed ASCII DOS console, which was resolved with the dip switches available on the card.
3. From the 3 Cirrus Logic video cards the STB one was most reliable, also, it produced better picture than the other two. Picture quality was on par with the ET4000/W32i card.
ET4000 produces "paper"-like warm colors that are very easy on the eyes, while the STB's color temperature is "cool" and more crisp.

THE CULPRITS

BENCHMARKS

As i said above - tests were performed on a late ASUS VLI-486SV2GX4 board with DX5-160MHz CPU.
Shown below are Doom and WinTune2 results only.

CONCLUSION
Cirrus Logic GD-5434 based cards are the fastest ISA graphics adapters, but it takes a fast 486, or later, computer to show the difference.
On a 386 class hardware there is virtually no performance difference between the different cards - performance is entirely CPU bound.
The image quality of the Diamond card was overall better. Both STB and Diamond cards behaved equally well in all tests.

Btw, the late ET4000/W32i model seems to be one of the best VGA ISA adapters i have seen to date - solid, fast and with great image quality. Takes with ease whatever motherboard is thrown at it.

386 motherboard JUKO 386DX W/B based on OPTi 495SX.
An interesting and great looking piece of hardware from year 1993.

Board/chipset is very versatile.
Offers both clock generator and crystal oscillator circuitry.
Covers wide range of CPUs - from 386DX to 486DX#.
Minimum supported frequency is 33MHz, maximum is 80MHz assuming 486 CPU, but i am sure it can go higher than that.

There are least 3 different ISA variations of the motherboard and one VLB one - some are with soldered CPU and no socket but with available through-holes, others don't have jumpers for Cyrix CPUs, third don't have clock generator socket + related jumpers.
This one here uses crystal oscillator, which I prefer, because it is more flexible for overclocking than the few predefined frequencies of the clock generator.

Very stable - runs 386DX and Ti486SXL2 CPUs at 55MHz, as well as IBM BL3 at 100/110 MHz.
And the best part - it just works !
The only stability flaw: if SXL2 CPU is used and level 1 cache is enabled the CF cards become problematic. Technically they work, but data corruption can occur.
Apparently not all IDE controllers are equal in this regard. Promise EIDE Pro and UM8673F based ones held well. Took a moment to figure this out.

The board is not pretentious about memory and video cards. Takes with ease any decently good RAM and VGA adapters.
This is nicer compared to the DTK PEM system for example, which gets picky about RAM and VGAs when running at 50MHz+ with the TI SXL2-50 and IBM BL3 processors.

Sporting Ti486SXL2-50 CPU, 32Mb 60ns RAM, 256Kb 12ns SRAM.
Squeaky clean. Wonder if it was used at all ...

While things tick well at 55MHz, there are better clock-to-clock performers out there, but with the benefit of extra frequency this fellow does just fine.

Tested with STB Nitro ISA (Cirrus Logic GD5434) video card.

--- AMD 386DX at 55MHz, ISA at 13.75MHz

All BIOS settings on max except:
CACHE READ CYCLE = 3-1-1-1 (best is 2-1-1-1)
CACHE WRITE STATE = 1 (best is 0)

--- TI 486SXL2 at 55MHz, ISA at 13.75MHz

All BIOS settings on max except:
CACHE READ CYCLE = 3-1-1-1 (best is 2-1-1-1)
CACHE WRITE STATE = 1 (best is 0)

Speedsys hangs when L1 cache is enabled.

--- IBM BL3 at 100MHz (2x50), ISA at 12.5MHz

All BIOS settings on max for DOS.
For Windows CACHE READ CYCLE = 3-1-1-1.

Most FPUs have problem in this configuration.
The only one that holds well is ULSI DX/DLC 40MHz.

--- IBM BL3 at 110MHz (2x55), ISA at 13.75MHz

All BIOS settings on max except:
CACHE READ CYCLE = 3-1-1-1 (best is 2-1-1-1)
CACHE WRITE STATE = 1 (best is 0)

System is not fully stable, but most DOS games and utils seem to be functioning.

---

benchmark results

A really nice motherboard that just works.
If one wants to build a classic high-end 386 system - this is a good place to start.

A while ago @ekb mentioned to me that the 286 VLSI system from page 1 may not be long-term stable, despite my initial claim for complete stability.
He has the same board and was having hard time stabilizing it to survive 10 or more rounds of CheckIt3 memory test.
I tried that here and yes - got memory failures too. Bummer.

This sent me on a new "hunt" for good SIMMs.
Over time distilled a handful of them that where almost stable.
Interestingly enough they were not all 60ns rated chips, but a mixture of different brands and rates.
Those got me to the 90% mark, but no further.
Along the way i realized that there is a thermal problem at play that needs to be solved to hit the 100% stability.
Apparently this motherboard pushes the memory chips to the edge and at the moment they get warm - errors occur.

Btw, i have 2 more 286 VLSI NEAT mobos - one with the same #201 chipset as the one in question here and another one with #101 chips.
They both have the same issue. Anyway.

Experimented with several designs for ventilation and cooling.
Goal was to improve thermal conditions, yet not turn the case into a helicopter.

Ended up with a 3D printed holder for two turbines over the memory slots + bigger fan at the top of the case:

The "contraption" in action after 10 rounds of CheckIt3:
https://www.petershipkov.com/temp/retro_pc_im … m_stability.mp4

Phone's microphone amplifies the noise from the PSU's fan enormously. 😖

That's quite the collection of retro hardware you've got there! Really nice builds and splendid work with all the photographs and background info. Must have taken countless hours and hours to do all this.

Thanks Hezus.

Btw i actually do not collect retro hardware. Hoarding is not my thing.
I am more into building handful, but really good (to my liking at least), fully functional retro PCs.

As for this "blog" here - out of interest i acquire specific items - things that i could only dream about back in the day - play with them for a while and then flip them, but at a lower price.
I am no Mother Teresa, but trying to be mindful of the community and not inflate further the already ridiculous prices scalpers ask for old "junk".
I keep around some of the cooler toys, but they eventually go too.
Anyhow. 😀

-----------------------------------

Leadman Electronics Inc. LM-103FS

Speaking of which.
For a while i wanted to check the latest and greatest of the 286 Headland chipsets - HT12P-16/A, so finally found a motherboard based on it.
Standard late 1990+ layout.
High-quality, well preserved.
The only downside was the lack of CPU socket.
It came with a 16MHz Harris CPU that couldn't go above 20-ish MHz.

So i decided to swap it with a proper 25MHz rated processor ... and almost destroyed the board along the way.
My soldering skillz are pretty ok, so i didn't expect much trouble, but after "bolting" the CPU on, the board fell silent.
Inspected my work multiple times with multimeter and optically - all was good.
Then i realized that the CPU is oriented the wrong way. No idea how i ended up doing that, after all the careful planning and stuff.
Unsolder, prep the board pads and cpu pins, solder again - still silence. What the heck ...
Unsolder, put the original 16MHz CPU - it works. Hmm ...
Solder the 25MHz one - silence.
During this excercise the integrity of few CPU solder pads on the motherboard started to deteorate. 3 of them fell ...

Then i came with the bright idea to check if the 25MHz CPU actually works - tried it on another board - silence.
I am not sure why i didn't test it beforehand, too late now. In the thrash, pulled another one from the vault, tested it - works. Ok.

Close inspection of the solder pads didn't bring any confidence. It was clear that i have only one shot left before losing the board.
Threw everything i had at repairing the 3 missing and 2 loose ones. Turned out pretty well. But even i used these special tin steel wires, the added height was just enough for many CPU pins not to contact well enough the pads underneath.
Heatgun+paste wouldn't be sufficient. Hand soldering for the win.
I was ready for this not to work at the end, but it actually turned really well.
Hollywood style happy ending.
First world drama !

Effort was worth it.
Motherboard is really good.
Supports shadow ram and ticks well at 25MHz.
Does not offer control over wait-states, so it cannot go above the 25, but still.
As all boards at that frequency it gets picky about RAM chips quality.
For long term stability SIMMs have to be ventilated.

The usual set of stats and benchmarks.

benchmark results

Conclusion:
Despite the initial setbacks, the end result was pretty good - fast and reliable motherboard/chipset.
Played few games for a while, ran some 286 grade computation tasks - everything works as advertised.
I often associate Headland with "average" and "boring", but this one is different.
Looking at the scores - results are in the ballpark of the VLSI boss.

Alaris Cougar II (Cobalt-AT 486BL) based on OPTi Premium 82C499 B
Comes with soldered on-board IBM 486BL3 (blue lighting) processor, 486 Socket 2 (Intel Overdrive ready), Appian ADI/2 EIDE controller (Adaptec AVA-2625VL).

An interesting and rare piece of retro hardware.
Both boards are really well preserved.

Came with 256Kb 20ns SRAM. Upgraded to the maximum 512Kb with 10ns rated chips.
Installed 64Mb of 60ns RAM (Micron). Downgraded to 16Mb since the IBM BL3 chips can cache only that amount.
Added 33MHz rated Cyrix FasMath FPU (gray top - the fastest core).
Goal is as usual - fully stable, maxed-out system.
Easily achievable at 75MHz.

The installed CPU heatsink is totally insufficient.
There are thermal issues even at stock 75MHz.
The CPU can hang if pushed hard enough.
Removed it and strapped a properly sized heatsink+fan.

The Adaptec VLB AVA controller is very capable.
No complains there whatsoever.

Never liked Mr.BIOS much as it often tries too hard and that sometimes gets in the way.
For example - after frequency change (using jumpers) i have to go and correct BIOS settings, because it auto-adjusts them according to some internal lookup table. Mildly annoying. But other than that this BIOS implementation is actually pretty ok.

User Twinge provided BIOS version with unlocked multipliers here.

Two obvious differences:
- no sticker (oh, no!)
- FPU socket lid pointing south-east while the first pin is actually located south-west.
One similarity:
- identical performance and stability

--- EDIT
Obtained the two boards below much later than the two above.
They can operate stably at 100MHz (2x50, modified BIOS) on air cooling which is impressive and will be covered in a separate post later.

This one handles all BIOS settings on max, except LEVEL 2 CACHE WRITE = 1 WS (best is 0) otherwise POST hangs.
Soft reset does not work.

This one is flawless - a zero wait-states all BIOS settings on max system.
Well, with one exception, as with the board mentioned above, the soft reset does not work.

--- END EDIT

While everything works really well at 75MHz (3x25MHz) with optimal bios settings and 0-wait-states jumpers configuration, the system is not fully stable at 3x33MHz.
This can be overlooked if one runs basic synthetic tests, or performs some casual activities, but Doom demo runs often quit early, 3D rendering tests hang, etc.
Without getting into the details - i tried hard to overcome the instabilities, but without success.
It is difficult of course to pinpoint where the problem is by just observing the side effects of it - the hangs.
The used sram, ram and vga components are of high quality and can withstand much more extreme conditions than these 33MHz.
I think the CPU can run well at 100MHz, but there seems to be a problem somewhere between the L1/L2 caches. If either one is disabled - everything is fully stable, but at this point kiss any performance good bye ...
For now I am going to blame it on the motherboard.

75MHz

100MHz, system is not fully stable (yet)

Video showing 100MHz run with Ark1000 VLB video card
While Ark1000 delivers the fastest DOS graphics, there were some artefacts.
S3 Trio64 VLB produced the best picture, but at the expense of small performance drop

benchmark results

This system offers a lot for DOS gaming and casual activities, but the low FSB/FPU frequency are insufficient for serious retro computing (wait, what ?).
So, if you are one of the two people out there who is into folding of proteins on a 386 class hardware - better look elsewhere.
Joking aside - depends on the point of view, Cougar+BL3 can be considered the pinnacle of all things 386 ... or the Quasimodo.

---

Evolutionary step forward using better BL3 CPU that enables 100MHz frequency at much improved stability. Described here.
It will take active cooling with Peltier to bring stability under complete control.
Need to think how to handle that given the surface mounted CPU and limited space around it.