Octek Fox-286 with Headland chipset.

Octek had couple of hits with their hardware back then.
286 Headland chipsets were very common ... and just as much unexciting.
So, i was not sure what to expect here.

The board is of good quality. Had to solder a crystal oscillator socket.
Takes up to 4 Mb of RAM. Used 4x1 60ns sticks.
DIP switches provide control over 0/1 wait-state and RAM configurations.

At 0 wait-state 16 MHz is the limit. Above that the board does not light-up.
At 1 wait-state things go as high as 27.5 MHz, fully stable.

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

The usual set of stats and benchmarks:

Conclusion:
An average board without unique features.
In search of positive comment i am arriving at - 30 years later it still works.

Something different.

I have been slowly setting up a clean 1994 486 VLB system, but had some doubts about the casing.
Never owned branded PCs. Custom builds all the way, tuned for maximum performance. All bells and whistles removed - only the essentials left.
The standard metal cases were always a point of annoyance. Lids were rarely closed, many internals removed for convenient access to hardware.
So, when the retro bug bit me a while ago, i started looking for a proper alternative to the big-bulky-clumsy cases - i have a fridge in the kitchen already. 😀
At the end i decided to make a 3D printed PC case to my liking: austere and functional.

Requirements:
- simple shapes that can be easily augmented into other designs
- easy to print objects
- economic on filament - fit within 1 and a half spools, including some reprints
- structural strength
- easy to assemble
- functional and easy to use
- compact

Ok, here goes nothing:

As i said - design is minimalistic.
I stopped using HDDs and moved to CF cards long ago, so there is no holder for HDD in there, but space is allocated on the side of the PSU separator, if one day i decide to go back to HDDs.
The design allows to easily add/remove the PSU. Reason is - i have more systems than PSUs. When want to fire-up one of them i grab an available PSU and play.
Ventilation is secured by the PSU - sucking "hot" air from the inside and expelling it.

There are countless brands and types of 3D filaments. Most of them are not great, but there are few that are just right.
My 3D filament of choice is Polymaker PolyLite PLA 3D, 1.75mm, temperature 220/70:
https://www.amazon.com/gp/product/B01IAVQIWE/ … 0?ie=UTF8&psc=1
It is the easiest thing to print with using Prusa I3 MK2 (and later models) - precise clean shapes, does not bend at all, sticks really well - you can literally print on a cookie and it will be fine.
My first pick was natural transparent filament, but wife said "looks old at first sight", the printing quality was superb.
Black looks cool, but absorbs too much light and the case gets too dark on the inside. This compromises the requirement for "functional", so ended up going with white filament. Next time when I drop a screw by accident inside the case it will not be lost for good 3 minutes.
Anyhow.

Everything needed for the case - bunch of printed parts, couple of screws, "reset" cable, Loctite glue.

The "reset" button consists of 4 pieces + cable. The contact surfaces that will be glued together required a bit of smoothing - scraped them few times with a knife.

The slot for the PSU's power switch. Took some trial and error to make a design that allows easy insertion and removal of the switch, but keeps it tightly fit while inserted there.

The individual panels glued and drying for few minutes.

This is perhaps the most tricky part of the whole assembly - the screw sockets for the motherboard.
I have printed markings on the bottom panels indicating where the slots should be glued, but wanted to double check in case my blueprints were off.
Inserted the back panel into the corresponding slots and placed the sockets.

Put the motherboard on the screw sockets. The two inserted cards helped me position the mobo precisely in the right place.
Notice the green sharpie with extended graphite.

Used the sharpie to mark the places of the sockets. Turned-out my printed markings were correct.

Test assembly - inside. Don't pay attention to the messy cables, this will be sorted out later on.

Test assembly - outside. Didn't spend time to completely fit all panels, that's why they look a bit dislocated, but In fact the panels fit and hold onto each other so well that the case can be used like that without further gluing.

In case any of this is interesting to you - here is the blueprints (Maya ASCII format), triangulated OBJ files, factory file (Simplify3D), gcode files ready to print:
https://www.petershipkov.com/temp/retro_pc_im … _at_pc_case.zip

Leaving the final assembly for the next post, when the 486 system is in complete order.

That' really cool. Love it.

I like my 486DX5 PCI/VLB/ISA PC from page 1 a lot, but my purist nature has been growing on me and i ended up building a clean VLB system from year 1994.

HARDWARE

• motherboard: ASUS VLI-486SV2GX4 rev 2.1, 1024Kb 10ns L2 cache
  cpu: Am486DX4-100V8T 120MHz
  • AMD-X5-133ADZ 160MHz
  ram: 64Mb 60ns parity FPM (Micron - the picture below is old and shows 2x16Mb, instead the current setup with 4x16Mb)
  vga: ARK1000VL 2Mb (primary)
  • Diamond Stealth 64 DRAM 2Mb (S3 Trio64)
  scsi: Adaptec AHA-2840/42A + UMC multi I/O card for ports
  ide: Promise EIDE2300 Plus
  audio: Sound Blaster 16 CT2230
  lan: 3com EtherLink III
  psu: 450W AT
  input: BTC 5121, MS IntelliMouse 1.1A
  os: DOS 6.20, Windows 3.1
  partitions: 1x504Mb

NOTES

This build was less straightforward than expected for 3 reasons:

Initially i wanted this to be an Intel CPU based system, but had to switch to AMD to achieve complete stability at 120 MHz.
Intel's 486DX4-100 CPU running at 120 MHz was fine for the most part, but failed the 3D rendering tests.

The situation with HDD controllers was more hairy.
Obviously i couldn't go with ISA SCSI/IDE controllers for this rig, so i had to choose between SCSI Adaptec AHA-2840A and bunch of VLB IDE/EIDE ones.
Adaptec AHA-2842A operates in 10 Mb/s setting and achieves ~7.2 Mb/s read and 4Mb write speeds - not bad, but some of the EIDE adapters around that time spank it on performance. Also, the problem with this guy is that it uses bus mastering. Things work well with the DX4 CPUs because they do not use WB1 mode, but if i decide to switch to AMD DX5 CPU for example, i need to either disable the WB1 cache with jumper/bios, or use an EIDE adapter.
I wanted to find the best option, so i spent a moment to test couple of VLB IDE/EIDE boards. This revealed interested details that i will post at some point later. The best performing and most stable EIDE VLB adapter was Promise EIDE2300 Plus, so it was given the honor to complement the rig. Driver in DMA mode, speed 8 (highest). It is worth mentioning that it works just fine in DMA mode with POD CPUs + L1 cache in WB mode - something not all VLB EIDE controllers can brag about.

I had to make some "difficult" decisions about VGA adapters.
There are quite a few good ones in the pack.
In Windows things are very simple - S3 Trio64 then the rest. In 16 bit mode it starts to even take on Matrox Millennium PCI.
In DOS Ark1000VL is the fastest thing around followed an inch behind by S3 Trio64.
At the end i went with Ark1000VL, because this is more or less a DOS PC. Other than few Win3.1 apps - there is little else for me there.
Will provide in subsequent post the collected performance information about VLB/PCI VGA adapters.

Getting the system maxed-out with 64Mb of RAM and 1024Kb of SRAM + WB2 mode was a real challenge.
I had to comb through a lot of cache chips to find 9 that work well.
So far so good, but i have this nagging suspicion that when in WB2 mode things may not be solid 100% stable.
Looking for ways to prove that ...
EDIT: Few months passed already and the system is just solid. Ok, i was not able to prove it unstable, but gladly accept the result.

There are mild differences between revision 2.1 and 2.0 in the upper right corner, related to CPU voltage control.

Will not post pictures of the rest of the VLB adapters here, they are already shown on the first page of this thread.

486DX4 at 120MHz with Adaptec AHA-2842A

How things look in the physical world:

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

Tested the system with AMD-X5-133ADZ running at 160MHz with Promise EIDE2300 Plus and Transcend 133x CF card, which is highly compatible with this class hardware (no write access issues and other nonsense).

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

I couldn't resist to check how this clean VLB system compares against two other champs:
Biostar MB-8433UUD-A (ISA/PCI) - one of the fastest 486 PCI boards in terms of VGA performance (despite some stability issues, 3D rendering, etc.)
Asus PVI-486SP3 (ISA/VLB/PCI) - what i consider the Maybach of 486. 😀

Matrox Millennium seems to be the fastest PCI card until the end of year 1996, so i picked it for this comparison.
Used Ark1000VL as a VLB representative, included S3 Trio64 for the Windows test only.

Linking here the composed benchmark results as well.

VESA local bus video cards have clear advantage in DOS, but lack in Windows.
As i mentioned above already, will try to post the full stack of VLB/PCI VGA perf data in the coming days.

Asus PVI-486SP does really good job at squeezing the most out of the CPU - the best i have seen among 486 mobos.
So i am not surprised that it outguns the VLI board at the 3D rendering tests by quite a margin.

At the end of the day, i am still flip-flopping with this rig between 1994 486DX4@120MHz+SCSI and 1995DX5@160+EIDE.
It feels like it will be an 1995 rig at the end of it, which means that i will need to get back to making a 1994 one at some point later - not a bad thing to look forward to. 😀

The Asus VLI-486SV2GX4 story evolved with AMD Am5x86 processor running at 200MHz (4x50)

Very nice job, impressive.

For best performance with your SV2GX4 you should activate the dirty bit in your BIOS in L2WB. Therefore you have to modify some registers and update your BIOS. The official BIOS doesn’t support dirty bit.
The Am5x86 runs well with L1WB together with the AHA2840 (2842 is the version with Floppy). Make sure you run the newest BIOS on your board and close the Jumper J5 on your 2840.

A pro tip.
I remember reading with great interest your thread Evolution of a Socket3 System to a POD @100MHz about that stuff.
I use the BIOS you mentioned already, but at the moment J21's pins 4-5 get shortened for WB on the motherboard all bets are off for IDE bus mastering, regardless if J5 is open or closed on the SCSI controller. Or do i miss something here ? Let me know.
Thanks.

Try the Jumperset for P24D and add the following jumpers:
JP16: 1-2 (important for the CPU Voltage!)
JP20: 1-2 (multiplier 4x)

Always close the J5 on the 2840.

Try to toggle JP26 between Delay and Non Delay
Try to toggle between the VLB slots

I got L1WB stable on my SV2GX4 without DMA problems with the following CPUs: Am5x86 (BGC), AMD DX4 SV8B, Cyrix 5x86, intel DX2 66 (P24D)
With DMA problems (Floppy and/or SCSI access) or not working with L1WB: Pentium Overdrive (P24T), intel DX4 WB (SK096)

The 3D printed case is very impressive. I wouldn't have been able to design something like that, at least not currently! I also just bought a Matrox Millenium for the same reason. Will continue to watch this thread with interest.

@PC-engineer
This is how I have the jumpers already.
I bet it is an Acard scsi2ide adapter issue (stopped using HDDs long ago).
Will strap a scsi drive to verify.

@pii_legacy
Yeah, the case turned out pretty ok. Glad to see you like it.
Matrox Millennium has two flavors. The main chip on some of them ends with R2 others have R3. Usually the ones with R3 have the latest BIOS which makes a small positive difference for performance in some cases. You can also upgrade bios manually, but the DOS utils for that are not great.

It looks like this is Acard scsi2ide related issue that is not present when using SCSI HDDs.
For now i will let it slide, since the problem does not affect the WT AMD 486DX4 CPU in use.

--- premise

While assembling the clean 486 PCs from my previous posts, i spent the time to check how VLB and PCI video cards from years 1993-1996 compared against each other. Wanted to see what the progression from early VLB to early PCI cards looked like.

--- test configurations and related notes

Performed an all-to-all test with the 3 best 486 motherboards for 160MHz (4x40) configuration:
- Asus VLI-486SV2GX4 rev.2.1, 1Mb 10ns L2 cache , 64Mb 60ns FPM RAM
- Asus PVI-486SP3 rev.1.22, 512Kb 10ns L2 cache, 64Mb 60ns FPM RAM
- Biostar MB-8433UUD-A rev.3, 512Kb 10ns L2 cache, 64Mb 50ns EDO RAM

Default hardware/software setup:
- all BIOS settings on max
- VLB wait state = 0
- VLB speed <= 33MHz

VLB cards were tested with Windows 3.11, whereas the PCI cards were tested on Windows 95 OSR2.
The primary consideration was that some VLB cards lack (good) drivers for Windows 95, and similarly, some PCI cards don't have (good) drivers for Windows 3.11.
The resolution of 800 x 600 pixels with 16 bits per pixel color depth serves as the primary benchmark. This configuration is chosen because it represents a common denominator supported by nearly all cards in this category.
For some of the more advanced VLB graphics adapters i conducted a wider range of tests at higher resolutions and color depths. However, to maintain clarity and prevent the table below from becoming completely unwieldy, I've chosen to share only the 800 x 600 x 24 bpp results as they effectively represent the observed performance characteristics.

If multiple drivers were available, the fastest one was used. That specific version is listed for each card.

Various BIOS versions, including those from different brands, were evaluated for the more noteworthy graphics cards with no performance difference observed.
A comprehensive cross-comparison of all available BIOSes for all cards would be a time-consuming endeavor and has been deferred for future consideration.

In the realm of these vintage graphics adapters, the physical form factor of the RAM chips - SOJ or DIP - does not impact performance or stability.
While cards equipped with SOJ chips may appear more modern, this is purely a cosmetic distinction.

Multiple brands competed in the market space during this era, each offering an array of models.
While testing every single one of them is impossible, considerable effort was invested in selecting appropriate candidates for evaluation based on the following criteria:
- cards from reputable manufacturers
- well-preserved hardware assemblies
- minimum of 2MB RAM, particularly crucial for VLB models
- preference for VRAM and FPM DRAM over standard DRAM

To avoid redundancy, only one representative card per graphics chip design is included at this time. Exceptions can be made for:
- interesting variations between the different models, such as memory type
- drastically different performance/stability

--- VLB video cards

A complete list of all graphics chips that were used in commercially available VLB video cards.

(Aliance ProMotion 3210) Miro VIDEO 12PD VL, 2Mb
(ARK Logic ARK1000PV) Hercules Stingray, 2Mb
(ARK Logic ARK1000VL) unknown brand/model, 2Mb / 2TheMax Premier 1000, 2Mb
(ATI Mach64) ATI Mach64 Graphics Pro Turbo, 4Mb
(ATI Mach64) ATI Mach64 Graphics Xpression, 2Mb
(ATI Mach32) ATI Mach32 Ultra Pro, 2Mb
(ATI Mach32) ATI Mach32 Graphics Wonder, 2Mb
(Avance Logic ALG2228.A) Avance Logic VGA228-VT REV:A, 2Mb
(Avance Logic ALG2228) ExpertColor DSP-2228, 2Mb
(Avance Logic ALG2201.0F) ALi AV100-VL Rev.2, 2Mb
(Chips F64300) Chips and Technologies KC7CT-6400, 2Mb
(Cirrus Logic GD-5434) Orchid Kelvin 64-VLB, 2Mb
(Cirrus Logic GD-5430) Orchid Kelvin EZ-VLB, 2Mb
(Cirrus Logic GD-5429) Quantum Designs CL542X/SOJ/SMT REV 1.0, 2Mb
(Cirrus Logic GD-5428) FIC Combo-542VP, 1Mb / PALM VL-BUS revision B.3, 2Mb
(Cirrus Logic GD-5426) Genoa Windows VGA 24 8520VL, 2Mb
(Cirrus Logic GD-5424) Protac 826W, 1Mb
(IIT XGA015) Hercules Graphite VL Pro, 2Mb
(Matrox IS-ATLAS R1) Matrox Ultima 503-AA REV A, 2Mb
? (Matrox IS-Titan/Ultima-Plus)
? (Matrox IT-Dubic/Impression)
? (NCR 77C22E)
(OAK OTI087) OAK OTI-087 VL BUS VER 1.0, 2Mb / OAK V0874003 REV:A, 2Mb
(S3 Trio64V+ 86C765) Madao’s MK-765VL-000, 2Mb
(S3 Trio64 86C764X) Diamond Stealth 64 DRAM T VLB REV B2, 2Mb
(S3 Trio64 86C764-P) Diamond Stealth 64 DRAM T VLB REV B2, 2Mb
(S3 Trio32 86C732-P) Diamond Stealth 64 SE VLB REV A1, 2Mb
(S3 Vision968 86C968-P) Number Nine 9FX Motion 771 VL, 4Mb
(S3 Vision964 86C964-P) Diamond Stealth 64 VRAM VLB REV A6, 4Mb
(S3 Vision868 86C868-P) Diamond Stealth 64 DRAM VLB REV A3, 2Mb
(S3 Vision864 86C864-P) Number Nine GXE64, 2Mb
(S3 928 P86C928G) VideoLogic 928Movie 99HG-3.10/3.12, 2Mb
(S3 928 86C928-P) Diamond Stealth PRO VL REV C3.1, 2Mb / ELSA XHR Winner 1000 VL, 3Mb
(S3 928 P86C928) Diamond Stealth PRO VL REV C3.1, 2Mb
(S3 911 P86C911) MSI MS-3401, 1Mb (note - this is not a VLB adapter but Micronics local bus one)
(S3 805i P86C805i-P) Jaton VL41C/V5, 2Mb
(S3 805i P86C805i) Miro CRYSTAL 16Si/VL, 2Mb
(S3 805 P86C805-Q) Soyo DSP3805B VER:1.0, 2Mb
(S3 805 P86C805-P) Quantum Designs S3805VLVGA, 2Mb
(S3 805 P86C805) Orchid Fahrenheit 1280 Plus/VLB, 2Mb
(Trident TGUI9440-1) Jaton VL46F-R4, 2Mb
(Trident TGUI9440AGi) Jaton VL46A/V2, 2Mb
(Trident TGUI9440) Quantum Designs PB-TD94540VL/SOJ/SMT/V3, 2Mb
(Trident TGUI9420DGi) Jace Technology JC9224, 2Mb
(Trident TGUI9400CXi) Trident 7343L REV 6A, 2Mb
(Trident TGUI9200CXr) Trident PTI-92CXR V0, 2Mb
(Tseng Labs ET4000/W32p) Gainward ET4000/W32P VLB 2.1, 2Mb
(Tseng Labs ET4000/W32i) BCM GP-4530, 2Mb
(Tseng Labs ET4000/W32) Machspeed VGA GUI 2400 / ET4W32VL-4.1, 1Mb
(Tseng Labs ET4000AX) Gainward 9207-02, 1Mb
(UMC UM85C418F-GP) Puretek PT-2037, 2Mb
(UMC UM85C418F-GP) TK TK85C418G-4J-D08, 1Mb
(WDC WD90C33-ZZ) Paradise VLBUS3300, 2Mb
(Weitek Power 9100) Diamond Viper SE VLB 2+2 Rev.A2, 4Mb (2Mb framebuffer, 2Mb for applications)
(Weitek Power 9000, Acumos AVGA2) Diamond Viper REV A7, 2Mb, 512Mb
(Weitek Power 9000, Oak OTI087X) Diamond Viper REV E4, 2Mb, 512Kb
(Weitek Power 9000, Oak OTI087X) Diamond Viper REV E3, 2Mb, 512Kb

--- PCI video cards

PCI cards extend well into the 2000s.
Limited the selection to year 1996 only.
Later variants don't make sense for 486 class hardware.

(Matrox MGA-2064W-R3) Matrox Millennium 590-05 REV B, 8Mb
(S3 Trio64V2/DX 86C775) Power Color C64+/V2, 2Mb
(S3 Virge/GX 86C385) Compaq, 2Mb
(Tseng Labs ET6000) Jaton Video Magic 128 8258A/V8, 4Mb
(Tseng Labs ET4000/W32p) Vision Magic ETW32PPS REV C, 2Mb
(Cirrus Logic GD-5480) Prolink MVGA-CL548XP Rev 2B, 4Mb
(Chips B65554) unknown brand/model, 4Mb
(Trident ProVidia9685) Union UTD74, 4Mb
(ARK Logic ARK2000PV) Diamond Stealth 64 2001, 2Mb
(S3 Vision968 86C968-P) Diamond Stealth 64 VRAM PCI, 4Mb
(SiS 9202) unknown brand/model, 2Mb
Number Nine Imagine128 II, 4Mb

--- photo gallery with notes

(Aliance ProMotion 3210) Miro VIDEO 12PD VL, 2Mb
Supports memory interleaving.

(ARK Logic ARK1000PV) Hercules Stingray, 2Mb
Supports memory interleaving.
While intermediate at Windows GUI this is the absolute VGA blaster.
The high performance has its price - these cards can be problematic in some motherboards.
Both chip revisions (PV and VL) exhibit exactly the same performance, stability, compatibility.

(ARK Logic ARK1000VL) unknown brand/model, 2Mb / 2TheMax Premier 1000, 2Mb (new in box, printed manual)
Supports memory interleaving.

(ATI Mach64) ATI Mach64 Graphics Pro Turbo, 4Mb with memory module
Best results achieved with driver version 3.098.
Requires DRAM SPEED = SLOWER for some tests to pass successfully with Asus VLI-486SV2GX4 2.1.

(ATI Mach64) ATI Mach64 Graphics Xpression, 2Mb
Supports memory interleaving.
Best results achieved with driver version 3.098.
Requires DRAM SPEED = SLOWER for some tests to pass successfully with Asus VLI-486SV2GX4 2.1.

(ATI Mach32) ATI Mach32 Ultra Pro, 2Mb
Best results achieved with driver version 2.6-3.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(ATI Mach32) ATI Mach32 Graphics Wonder, 2Mb
Best results achieved with driver version 2.6(BBS).

(Avance Logic ALG2228.A) Avance Logic VGA228-VT REV:A, 2Mb

(Avance Logic ALG2228) ExpertColor DSP-2228, 2Mb
No memory interleaving.
Best results achieved with driver version 2.0. The later 2.3 version seems to have higher overhead.
Requires LOCAL BUS READY = SYNCHRONIZE for complete stability with Asus VLI-486SV2GX4 2.1.
The WinTune2 tests in Asus PVI-486SP3 requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability in Asus PVI-486SP3 1.22.

(Avance Logic ALG2201.0F) ALi AV100-VL Rev.2, 2Mb
No memory interleaving.
Unstable beyond 33MHz in Asus VLI-486SV2GX4 no matter what, but works just fine at 40MHz in Asus PVI-486SP3.
There is only one Windows 3.1 driver available for the 2201. Tends to be problematic - encountered some visual artefacts.
Quake 1 and WinTune2 tests required L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability in Asus PVI-486SP3 1.22.

(Chips F64300) Chips and Technologies KC7CT-6400, 2Mb
Supports memory interleaving.
Requires LOCAL BUS READY = SYNCHRONIZE for complete stability with Asus VLI-486SV2GX4 2.1.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(Cirrus Logic GD-5434) Orchid Kelvin 64-VLB, 2Mb
Supports memory interleaving.
Best results achieved with driver version 1.74.

(Cirrus Logic GD-5430) Orchid EZ-VLB, 2Mb
No memory interleaving.
Best results achieved with driver version 1.24.

(Cirrus Logic GD-5429) Quantum Designs CL542X/SOJ/SMT REV 1.0, 2Mb
No memory interleaving.
Best results achieved with driver version 1.5.

(Cirrus Logic GD-5428) FIC Combo-542VP, 1Mb / PALM VL-BUS revision B.3, 2Mb
No memory interleaving.
Best results achieved with driver version 1.5.
Despite installing a second megabyte of RAM (the unusual ZIP packages) in the FIC card it is not being recognized.
There is no manual available online. Tried blandly a lot of jumper configurations without success.
Can be a BIOS issue, but didn't try other BIOSes yet.
JV3 controls the VGA wait states. Picture shows it as set to 1 wait state.
J1, J2, J3 control the IDE wait states.

(Cirrus Logic GD-5426) Genoa Windows VGA 24 8520VL, 2Mb
No memory interleaving.
Best results achieved with driver version 1.5.
Actually i started testing with Typhoon Master Combo Revision D but it didn't handle FSB beyond 33MHz so moved to the Genoa card.

(Cirrus Logic GD-5424) Protac 826W, 1Mb
No memory interleaving.
Best results achieved with driver version 1.5.
Requires LOCAL BUS READY = SYNCHRONIZE for complete stability with Asus VLI-486SV2GX4 2.1.
Soldered SOJ sockets and temporary removed the ZIP chips, but no RAM is recognized in this case. Perhaps some of the surrounding elements must be reconfigured. Something i don't want to do (added damage to the card while testing for the right configuration).

(IIT XGA015) Hercules Graphite VL Pro, 2Mb.
New in box.


printed manual

(Matrox IS-ATLAS R1) Matrox Ultima 503-AA REV A, 2Mb
Best results achieved with driver version 2.21.
Requires CACHE BURST READ CYCLE = 3-1-1-1 and DRAM SPEED = SLOWER in Asus PVI-486SP3 1.22 for complete stability.

(OAK OTI087) OAK V0874003 REV:A, 2Mb
No trace of information available for this card online. Jumpers do seemingly nothing.
Looks like a later variant of the card below, if the more compact board layout and higher density memory chips are a factor. Yet, it is the more fussy card between the two and in this way the slower one.
Windows 3.1 GUI is a no go with 64k colors.
Requires DRAM SPEED = SLOWER in Asus VLI-486SV2GX 2.1 for complete stability in DOS.
Requires DRAM SPEED = SLOWER and CACHE BURST READ CYCLE = 3-1-1-1 in Asus PVi-486SP3 1.22 for complete stability in DOS.
For now will only list the performance results here:
Wolf3D: 75.4 fps (VLI), 75.7 fps (PVI)
Superscape: 62.5 fps (VLI), 62.5 fps (PVI)
PC-Player: 21.3 fps (VLI), 21.1 fps (PVI)
Doom: 37.3 fps (VLI), 36.1 fps (PVI)
Quake 1: 14.9 fps (VLI), 14.4 (PVI)
WinTune2: unstable (VLI), unstable (PVI)

(OAK OTI087) OAK OTI-087 VL BUS VER 1.0, 2Mb
No memory interleaving.
Best results achieved with driver version 3.0a. Linear addressing is available only for 256 color modes, otherwise memory paging is used (slower).
Very picky about RAM chips. Took a while to find the right combination of 16 pieces for 2Mb.
Requires DRAM SPEED = SLOWER for successful boot into DOS with Asus VLI-486SV2GX4 2.1.
There is a later revision of the card (above) which is profiled predominantly towards 486 class hardware, while this version is truly on the cross road between 386 and 486.

(S3 Trio64V+ 86C765) Madao’s MK-765VL-000, 2Mb
Supports memory interleaving.
It is important to use this S3 reference BIOS for peak performance. It is a monster. Much better than the STB PowerGraph 64 Video BIOS that Madao patched for 0 wait states. Equipped with it, the card is more or less on par with Ark1000VL in DOS interactive graphics and even beats it in some of the tests.
Best results achieved with S3 reference driver version 1.70.04.
Unfortunately the adapter didn't work in Asus PVI-486SP3 - no lights at all. Tried different BIOSes - no go. Feels like something on signal level.

(S3 Trio64 86C764X) Diamond Stealth 64 DRAM T VLB REV B2, 2Mb
Very late version from year 1996. With drivers/apps CD + cover.
Supports memory interleaving.
Best results achieved with driver Diamond Stealth 64 version 1.37.

(S3 Trio64 86C764-P) Diamond Stealth 64 DRAM T VLB REV B2, 2Mb
Supports memory interleaving.
Best results achieved with driver Diamond Stealth 64 version 1.37.
No performance difference to the model above.

(S3 Trio32 86C732-P) Diamond Stealth 64 SE VLB REV A1, 2Mb
No memory interleaving.
Best results achieved with driver Diamond Stealth 64 version 1.37.
One of the fastest cards for DOS interactive graphics, second to Ark1000VL only.
Requires ISA BUS CLOCK = 7.159MHz in Asus VLI-486SV2GX 2.1 for complete stability. This has no impact on performance.
Requires DRAM SPEED = SLOWER and CACHE BURST READ CYCLE = 2-2-2-2 in Asus PVi-486SP3 1.22 for complete stability in DOS.

(S3 Vision968 86C968-P) Number Nine 9FX Motion 771 VL, 4Mb (2 on the front, 2 on the back)
Supports memory interleaving.
Requires LOCAL BUS READY = SYNCHRONIZE for complete stability with Asus VLI-486SV2GX4 2.1.
Best results achieved with S3 reference driver version 1.70.04.

(S3 Vision964 86C964-P) Diamond Stealth 64 VRAM VLB REV A6, 4Mb
New in a box. With a secondary role as a VRAM chips stash.
Supports memory interleaving.
Best results achieved with driver Diamond Stealth 64 version 1.37.

(S3 Vision868 86C868-P) Diamond Stealth 64 DRAM VLB REV A3, 2Mb
Supports memory interleaving.
Best results achieved with driver Diamond Stealth 64 version 1.37.

(S3 Vision864 86C864-P) Number Nine GXE64, 2Mb
Supports memory interleaving.
Best results achieved with driver Diamond Stealth 64 version 1.37.

(S3 928 P86C928G) VideoLogic 928Movie 99HG-3.10/3.12, 2Mb
Supports memory interleaving.
Best results achieved with reference S3 driver version 2.41.
The VideoLogic PowerPlay32 chip gets very hot and needs a heatsink.

(S3 928 86C928-P) Diamond Stealth PRO VL REV C3.1, 2Mb / ELSA XHR Winner 1000 VL, 3Mb
Not sure what to think about the memory interleaving support of these video cards.
The above 86C928G has twice better performance in Windows GUI when 2Mb of RAM is installed.
The adapter from Diamond matches it with just 1Mb of RAM, but adding a second megabyte of memory does not improve performance at all.
At the same time the ELSA card does not go past ~8500 kilopixels per second. That is with 2Mb or RAM.
The Diamond adapter requires VIDEO SHADOW = DISABLED to function properly in both Asus VLI-486SV2GX4 2.1 and Asus PVI-486SP3 1.22, but the ELSA one is just fine with VIDEO SHADOW enabled.
The ELSA card requires CACHE BURST READ CYCLE = 2-2-2-2 in Asus PVi-486SP3 1.22 for complete stability.
Used Diamond Stealth Pro version 3.0 for Stealth Pro VL and ELSA's driver version 5.30 for Winner 1000.
Because of the significant differences in behavior decided to present both cards.

(S3 928 86C928) Diamond Stealth PRO VL REV C3.1, 2Mb
The older BIOS version 1.11 on this card tends to be more problematic than version 2.02 above - no lights on power-on in some cases and other general instabilities.
Otherwise performance is the same for the two Diamond video adapters.

(S3 911 P86C911) MSI MS-3401, 1Mb
This is not a VLB adapter but a Micronics Local Bus one.
The card does not work when inserted in a regular VLB slots - PSU shuts down.
Finally, S3 911 silicon does not have VLB support.

(S3 805i P86C805i-P) Jaton VL41C/V5, 2Mb
Supports memory interleaving.
Best results achieved with driver version 2.30 and 2.41 (same).

(S3 805i P86C805i) Miro CRYSTAL 16Si/VL, 2Mb
Supports memory interleaving.
Best results achieved with driver version 2.30 and 2.41 (same).
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(S3 805 P86C805-Q) Soyo DSP3805B VER:1.0, 2Mb
No memory interleaving.
Best results achieved with S3 reference driver version 2.41.
There is a bit of unlucky story here. Over the years two 805-Q cards died on me before i got the idea to test them properly. First was DEC unknown model, 1Mb, then Jaton VL41F/V2, 2Mb.
Hope the Soyo board does not follow their faith anytime soon.

(S3 805 P86C805-P) Quantum Designs S3805VLVGA, 2Mb
Supports memory interleaving.

(S3 805 P86C805) Orchid Fahrenheit 1280 Plus/VLB, 2Mb
No memory interleaving.
Best results achieved with driver version 2.30 and 2.41 (same).

(Trident TGUI9440-1) Jaton VL46F-R4, 2Mb
Supports memory interleaving.
Best results achieved with driver version UA6.4.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(Trident TGUI9440AGi) Jaton VL46A/V2, 2Mb
Supports memory interleaving.
Best results achieved with driver version UA6.4.

(Trident TGUI9440) Quantum Designs PB-TD94540VL/SOJ/SMT/V3, 2Mb
No memory interleaving.
Best results achieved with driver version UA6.4.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(Trident TGUI9420DGi) Jace Technology JC9224, 2Mb
Last letter in the chip ID is "i" which suggests memory interleaving, but results are the same with 1 and 2 Mb of RAM.
Best results achieved with driver version UG6.1.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.
With 2Mb of RAM installed the card may start corrupting text in Windows 3.1. Introducing a wait state using the jumper in the lower left corner mitigates the problem at the expense of lowered performance.

(Trident TGUI9400CXi) Trident 7343 REV C1HK, 2Mb
Last letter in the chip ID is "i" which suggests memory interleaving, but results are the same with 1 and 2 Mb of RAM.
Best results achieved with driver version UA6.4.
WinTune2 tests in Asus PVI-486SP3 1.22 were performed with L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability.
Trident produced another model - 7343L REV 6A with the same performance and other characteristics.

(Trident TGUI9200CXr) Trident PTI-92CXR V0, 2Mb
No memory interleaving.
Best results achieved with driver version C4.50S.
Driver is unstable in 800x600x64k mode. All other available modes work just fine.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(Tseng Labs ET4000/W32p) Gainward ET4000/W32P VLB 2.1, 2Mb
Supports memory interleaving.
Requires LOCAL BUS READY = SYNCHRONIZE for complete stability with Asus VLI-486SV2GX4 2.1.
The graphics card features four additional RAM sockets beyond those required for the standard 2MB configuration.
However, populating these extra sockets does not increase the card's usable memory beyond the 2MB limit despite ET4000/W32p's datasheet stating up to 4Mb of addressable space.

(Tseng Labs ET4000/W32i) BCM GP-4530, 2Mb
Supports memory interleaving.

(Tseng Labs ET4000/W32) Machspeed VGA GUI 2400 / ET4W32VL-4.1, 1Mb
No memory interleaving.
This version of the chip supports 1Mb of video memory only. The on-board soldered RAM chips must be removed for the unpopulated sockets to be used.
Requires LOCAL BUS READY = SYNCHRONIZE for complete stability with Asus VLI-486SV2GX4 2.1.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(Tseng Labs ET4000AX) Gainward 9207-02, 1Mb
No memory interleaving.

(UMC UM85C418F-GP) Puretek PT-2037, 2Mb
Supports memory interleaving.

(UMC UM85C418F-GP) TK TK85C418G-4J-D08, 1Mb
Decided to list a second card based on the same VPU for 2 reasons:
1. Layout and features - the green card has sockets for an additional megabyte of video memory which results in much better performance in Windwos GUI, while this one offers fully featured multi-io capabilities.
2. This particular card demonstrates superior performance compared to the one above. Initially, I hypothesized that the performance difference might be attributed to variations in their BIOSes. However, swapping their BIOS chips didn't change anything.
A key distinction between the two cards lies in their wait state configurations. This faster card features a dedicated wait state jumper, which is currently set to 0 for optimal performance while the green one probably operates with a mandatory wait state of 1.

(WDC WD90C33-ZZ) Paradise VLBUS3300, 2Mb
Supports memory interleaving.

(Weitek Power 9100) Diamond Viper SE VLB 2+2 Rev.A2, 4Mb (2Mb framebuffer, 2Mb for applications)
Cannot make this card to work in Windows 3.1.
Tried different drivers for the 9000/9100 series but result is always black screen "out of sync".

(Weitek Power 9000, Acumos AVGA2) Diamond Viper REV A7, 2Mb, 512Mb
Best results achieved with driver Diamond Viper version 2.0.
The AVGA2 chip is faster than OTI087X on a clock-to-clock basis.
Requires L2 CACHE BURST READ CYCLE = 2-2-2-2 (best is 2-1-1-1) for complete stability with Asus PVI-486SP3 1.22.

(Weitek Power 9000, Oak OTI087X) Diamond Viper REV E4, 2Mb, 512Kb
Best results achieved with driver Diamond Viper version 2.0.
Requires AT BUS CLOCK = 7.159MHz in Asus VLI-486SV2GX4 for the OAK OTI087X chip connected to the ISA bus to be functional. This has no impact on the Windows GUI test results.

(Weitek Power 9000, Oak OTI087X) Diamond Viper REV E3, 2Mb, 512Kb
Best results achieved with driver Diamond Viper version 2.0.
Requires AT BUS CLOCK = 7.159MHz in Asus VLI-486SV2GX4 for the OAK OTI087X chip connected to the ISA bus to be functional. This has no impact on the Windows GUI test results.

(Matrox MGA-2064W-R3) Matrox Millennium 590-05 REV B, 8Mb with memory module

(S3 Trio64V2/DX 86C775) Power Color C64+/V2, 2Mb

(S3 Virge/GX 86C385) Compaq, 2Mb
Comparison of Virge GX/DX/VX cards was performed here. They offer the same performance for 486 class hardware, so the GX variant listed here is representative for all of them.

(Tseng Labs ET6000) Jaton Video Magic 128 8258A/V8, 4Mb

(Tseng Labs ET4000/W32p) Vision Magic ETW32PPS REV C, 2Mb

(Cirrus Logic GD-5480) Prolink MVGA-CL548XP REV 2B, 4Mb

(Chips B65554) unknown brand/model, 4Mb

(Trident ProVidia9685) Union UTD74, 4Mb

(ARK Logic ARK2000PV) Diamond Stealth 64 2001, 2Mb

(S3 Vision968 86C968-P) Diamond Stealth 64 VRAM PCI, 4Mb with memory module

(SiS 9202) unknown brand/model, 2Mb

Number Nine Imagine128 II, 4Mb

--- results

--- conclusion

VLB video cards have advantage in DOS but many of them tend to struggle in Windows accelerated GUI compared to PCI graphics adapters from the same time period.
Ark1000VL is consistently the fastest graphics adapter in DOS.
Matrox Millennium PCI and S3 Vision968 VLB are the fastest cards in Windows accelerated GUI.

These tests are also an indication of how VLB, PCI, and VLB/PCI motherboards handle graphics.
Boards with bridged VLB/PCI interfaces tend to be slower and more pretentious than VLB or PCI only ones.

@PC-Engineer

Spent a moment today to check what is happening with this VLB SCSI + WB1 business.
Long story short - for this to work 3 things need to be in place:
- the WB-enabled BIOS, obviously
- despite using AMD DX5 CPU, jumpers have to be configured for P24T with J20 set to 1-2 for 4x multiplier = 160MHz
- because i use Adaptec AHA-2840A (the one without on-board floppy controller) an IDE card with FDD controller needs to be present otherwise the Adaptec guy won't boot the OS

#3 paused me for a moment 😀
Back when i bought the SCSI controller i had the option to get the 2842A version, but since i would need an I/O card for mouse which comes with HDD/FDD controllers as well, i opted for the slightly cheaper 2840A.

One less puzzle around here.

ok, so, upon closer inspection and profiling the outlined above configuration is not completelly stable.
it works most of the time - for example - Doom, quake 1, some apps, etc. seem to be fine, but wolf3d and dune2 can hang the system. i am sure there will be other cases too.
bummer.
rolling back - stability must be complete.

ran a search, there are 2-3 threads about this subject but kind of inconclusive - didnt get the sense that the configuration was throughly tested.

If you use more than one VLB card with 40MHz the transparent mode gets buggy. Also switching JP26 (D-ND) can help.
And i found out long term stability issues (sporadic crashes) with the 10ns ISSI 128k8 chips (several batches from several vendors) at clocks over 33MHz. The latencys didn’t matter, only time (>30min running) and clock (>33MHz). Double checked it with other boards with the same behavior. Also a active cooling of the chips didn’t help. In the end i switched to 512kb cache with 15ns 64k8 chips from UMC.

Thanks for the info PC-Engineer.

To confirm for the record:
Assuming system configuration: AMD BGC 160MHz WB1 CPU, 40MHz FSB, 1024Kb 10ns ISSI SRAM (had to go comb through bunch of chips to find the right ones), VLB VGA, tightest BIOS timings, transparent mode.
VLB IDE + WB1 = long term stable.
VLB SCSI + WB1 = unstable.

But since this system is targeting year 1994, it is using AMD DX4-100 WT1 CPU, which means that all of the above VLB SCSI+WB1 specific issues are not present.

You can use your Am5x86 as a DX4 100 with 3x 33MHz to „emulate“ the AMD DX4 100, so you can keep it installed and switch between the classes. The AMD DX4 100 arrived the marked in the end of 1994 the intel DX4 100 in Q1/94. The intel has some improvements over the AMD, one is the better compatibility for clock sensitive applications.

Maybe you can try a last test with your SCSI controller and 33MHz FSB. If it is stable than the root cause is the overloaded VLB @40MHz with two cards. In this case you can try another combination (switch the ports, or take other cards) of the both VLB cards. For an overclocked VLB it depends on several (small) factors to get it stable.

And the effect of L1WB is overrated for real 486 CPUs and a sufficient amount of L2WB cache. Maybe it is better to drive L1WT and accept the 0...1% performance loss and keep the nice SCSI controller.

Made a mistake in my last reply - the system is targeting year 1994 as stated in the initial post, but not 1993. Thanks for pointing out. Fixed it above.

I measured the WT1 vs WB1 perf diff carefully - it varies within 1-4%, depends on the app/game.
It totally makes sense to switch to WT1, as you suggested, and trade the few percents of compute/graphics performance loss for 2Mb/s disk speed gain, but the force of irrationality is strong with me - these few extra % are more valuable than the significant gain in hdd speed, so there goes the VLB SCSI. 😀

Good point about 33MHz FSP - it will be interesting to see what this does to it.

Did you have this project for ATX and mATX sizes?

I don't have a ATX based design, but the current one can be adjusted easily.
ATX PSUs dimensions tend to be more volatile than AT ones - this will dictate case size.
Do you have a particular PSU in mind ?
The only thing that will need a bit of attention is to replace the AT switch slot with something that fits an ATX one.

I am going to dump here info about few more motherboards and other stuff.

ASUS ISA-386U3Q

An old new stock. Squeaky clean, manual, wrapping, etc.

This seems to be a 1993 model. Visually the motherboard looks great - tight and well organized.
There are pretty much no jumpers on it. There is one for the FPU, 2 for the cache configuration and that's pretty much it.
The BIOS lacks any sophistication - there are 3 parameters related to performance that can be set to "fast" or "slow".

It is a bit picky about memory. Had to try few different sets to find the right ones for it.
However its biggest drawback is the inability to handle properly CF cards. Concerning.
As a workaround i strapped a SCSI adapter and all was fine from that point.
The SCSI operates at 6.7Mb/s, which is not exactly stunning for 50MHz FSB.

The board overclocks to 50MHz with FPU and things are stable for DOS gaming, but for 3D rendering and in Windows i had to step down to 45MHz for complete stability.

Didn't bother running LightWave3D rendering test. It takes about 3 hours to complete and won't reveal anything surprising.
benchmark results

Conclusion:
I had high hopes for this motherboard, but it didn't quite hit the mark.
Issues with RAM, issues with CF cards, performance is so-so.
Otherwise it is a real looker. Not sure if pictures give it a justice.