Gona wrote on 2026-01-28, 18:41:

Wow! I didn’t think that 35 FPS would be possible on a Socket 5 VLB Pentium 100 system.
FOREX chipsets are overprinted. For me the FOREX really looks like OPTi Pyton or OPTi Premium. These are better with VLB than PCI (but not reach a native PCI chipset with PCI card on P100). The only one I have seen (and tested) which is not OPTi, the DFI G586VPA with VIA VT82C535MV/531MV/505/406MV.
I was tempted to do a speed benchmark using a Powerleap PL-Pro/MMX 6.0 with a Tillamook 266, but these boards are not compatible with many cards.

Yeah at first I thought the Forex chipset was an Opti relabel as well but tbh now I'm not sure.
The number of pins on the Forex 58C601/58C602 chips do not match either the Opti Python or Premium ('Cobra') chipsets.
The older Opti 82C571/82C572 '486/Pentium writeback' chipset DOES have the same number of pins.
But the Forex 58C613 VLB to PCI bridge chip is definitely not the common Opti 82c822 bridge that came with all of their chipsets; for one thing it has only 160 pins instead of 208, and it also has its own unique PCI vendor/device ID.

Would love to see some Powerleap VLB results. That's a sought after piece of kit.
Been wanting to do the same with an Evergreen Spectra 400 interposer but this particular motherboard's BIOS is very bare as far as CPU support goes, won't even init Pentium MMX chips.
Resident BIOS patching guy chkcpu is looking at whether it can be added.

Beerfloat wrote on 2026-01-28, 19:24:

Yeah at first I thought the Forex chipset was an Opti relabel as well but tbh now I'm not sure. The number of pins on the Forex […]
Show full quote

Yeah at first I thought the Forex chipset was an Opti relabel as well but tbh now I'm not sure.
The number of pins on the Forex 58C601/58C602 chips do not match either the Opti Python or Premium ('Cobra') chipsets.
The older Opti 82C571/82C572 '486/Pentium writeback' chipset DOES have the same number of pins.
But the Forex 58C613 VLB to PCI bridge chip is definitely not the common Opti 82c822 bridge that came with all of their chipsets; for one thing it has only 160 pins instead of 208, and it also has its own unique PCI vendor/device ID.

Would love to see some Powerleap VLB results. That's a sought after piece of kit.
Been wanting to do the same with an Evergreen Spectra 400 interposer but this particular motherboard's BIOS is very bare as far as CPU support goes, won't even init Pentium MMX chips.
Resident BIOS patching guy chkcpu is looking at whether it can be added.

Amazing that the FOREX has unique PCI vendor/device ID!
I have sold my VIA Socket 5 VLB and Powerleap PL-Pro/MMX 6.0 to Putas for him 3D Blaster VLB tests.
None of the Pentium VLB boards boots AMD (or Cyrix) CPUs (even the AMD-K5 family microcode is missing).
I have asked Jan Steunebrink to try adding AMD to an OPTi AWARD bios, He has written first:

"Today I had some time to look into this MP039 Award BIOS.
A very interesting BIOS that looks very simular to an OPTi Socket 3 486
Award BIOS that I patched for Am5x86-133 support way back in 1999.
These early Socket 5 BIOSes where in fact modified 486 BIOSes with added
Pentium and chipset support.
I did mainly patches for 486 and Socket 7 BIOSes, so I never did a K6(+)
patch on such an early Pentium BIOS but I wil try. The main problem is zero
AMD support in these BIOSes, not even for the AMD K5.. "

Than I have sent a late OPTi Viper bios (with AMD microcode) too:

"Indeed an interesting socket 5 BIOS with more 486 than Pentium support. 😉
As I expected, anything more than basic Pentium support is missing. No MMX
and no AMD support. Consequently there are so many CPU support functions
missing, that adding AMD K6 support is practically impossible. Building a
new BIOS would be easier, but this is way beyond my expertise. 😉

Then the M519 BIOS. This BIOS has a 2A5UN identifier and is therefore for
the OPTi Viper-N chipset.
This Viper (OPTi 556/558N) BIOS will probably not work on the MP039. The
MP039 has an OPTi Python (546/822) or Cobra (596/822) chipset and needs a
BIOS with a 2A5UI identifier"

douglar wrote on 2026-01-27, 15:00:

My experience was that the DRAM Mach32 did ~6% better in DOOM.

My experience with ISA and PCI DRAM/VRAM mach32 cards is that the VRAM cards have been consistently faster with DOS tests:

Re: Wow ISA BUS insanely overclocked and running DOOM at 35 FPS with an ISA ATI Mach32

That said, it does appear to be chipset-dependent as the performance was near identical on a 430TX board:

Re: Wow ISA BUS insanely overclocked and running DOOM at 35 FPS with an ISA ATI Mach32

I have some Doom and FastDoom v1.1.1 benchmarks I randomly did one day with my PCI cards on my Gigabyte GA-5AX with a cache modded K6-2+-570 running at 6*100FSB, with 128MB of RAM and K6INIT:

I only have a VRAM VLB mach32 card, so I can't do any direct comparisons. I'll need to check the exact model numbers of the cards when I get a chance.

Gona wrote on 2026-01-26, 21:10:

I have benchmarked my VLB cards (36 pieces):
https://gona.mactar.hu/VLB_benchmarks/

Nice work!

I'm surprised at how low the WD90C33-based card ranks in your DOS benchmarks. I have the same card (VLBUS3300) and on my tests it has a performance roughly similar to ET4000/W32P or S3 Trio64 in DOOM and the usual DOS benchmarks. Are you sure you set the jumpers for fastest speed? The 0WS one in particular should make a difference.

Gona wrote on 2026-01-28, 19:55:

Amazing that the FOREX has unique PCI vendor/device ID! I have sold my VIA Socket 5 VLB and Powerleap PL-Pro/MMX 6.0 to Putas f […]
Show full quote

Amazing that the FOREX has unique PCI vendor/device ID!
I have sold my VIA Socket 5 VLB and Powerleap PL-Pro/MMX 6.0 to Putas for him 3D Blaster VLB tests.
None of the Pentium VLB boards boots AMD (or Cyrix) CPUs (even the AMD-K5 family microcode is missing).
I have asked Jan Steunebrink to try adding AMD to an OPTi AWARD bios, He has written first:

Thanks for sharing! I'm also very interested in the MP039 and to learn more about the differences between the Cobra and Python chipsets.
They must be very compatible considering there's that Supermicro P54VL-PCI with some individual boards out in the wild mixing the 82C546 with the 82C597.
Jan has mentioned the same problems in the Forex board BIOS but he said he's looking into it.
It'd be great if we could find some way to get CPUs > P54c it to at least boot on these Pentium VLB boards, even if we need to let an interposer handle the voltage, and let the likes of k6init finish configuration.

TheMobRules wrote on 2026-01-29, 03:55:

I'm surprised at how low the WD90C33-based card ranks in your DOS benchmarks. I have the same card (VLBUS3300) and on my tests it has a performance roughly similar to ET4000/W32P or S3 Trio64 in DOOM and the usual DOS benchmarks. Are you sure you set the jumpers for fastest speed? The 0WS one in particular should make a difference.

I have used my VLBUS3300 this way:
JP1 1-2 closed, others opened;
W1 2-3 closed;
W2 2-3 closed.
What was your jumper settings?

DEAT wrote on 2026-01-29, 01:24:

douglar wrote on 2026-01-27, 15:00:

My experience was that the DRAM Mach32 did ~6% better in DOOM.

My experience with ISA and PCI DRAM/VRAM mach32 cards is that the VRAM cards have been consistently faster with DOS tests:

Re: Wow ISA BUS insanely overclocked and running DOOM at 35 FPS with an ISA ATI Mach32

Interesting, but I suspect that the main performance limitation on the ISA cards is different than the performance limitation that affects the VLB cards.

The difference on the VLB benchmarks would involve memory write latency. DRAM wins there with lower latency writes. Lower write latency means the CPU can fling pixels faster, and when there is contention with the ramdac, the stalls are small because of the higher bus speed.

The difference on the ISA benchmarks involves the bus stalls. It doesn't take a lot of ISA bus stalls to overshadow the memory write latency difference. VRAM is going to have fewer bus stalls on pixel pushes because there is no contention with the ramdac.

Gona wrote on 2026-01-29, 16:38:

I have used my VLBUS3300 this way: JP1 1-2 closed, others opened; W1 2-3 closed; W2 2-3 closed. What was your jumper settings? […]
Show full quote

I have used my VLBUS3300 this way:
JP1 1-2 closed, others opened;
W1 2-3 closed;
W2 2-3 closed.
What was your jumper settings?

My settings are:

JP1 1-2 and 3-4 closed, 5-6 open
W1 2-3 closed
W2 2-3 closed

So it seems Jp1 3-4 is the only difference, give it a try and let me know! Documentation is scarce about this card so I cannot tell if that jumper is for 0WS...

TheMobRules wrote on 2026-01-29, 17:58:

My settings are: […]
Show full quote

My settings are:

JP1 1-2 and 3-4 closed, 5-6 open
W1 2-3 closed
W2 2-3 closed

So it seems Jp1 3-4 is the only difference, give it a try and let me know! Documentation is scarce about this card so I cannot tell if that jumper is for 0WS...

Thank you! I'm going to test on friday/saturday.

TheMobRules wrote on 2026-01-29, 17:58:

So it seems Jp1 3-4 is the only difference, give it a try and let me know! Documentation is scarce about this card so I cannot tell if that jumper is for 0WS...

Yes I think that's the local bus timing selector jumper. I have a card based on the same chipset - different layout but the jumpers all look the same:

https://theretroweb.com/expansioncards/s/west … elerator-vl-plu

And I've found the same thing - when the jumpers are set right it performs up there with the Tsengs, etc.

@Gona thanks so much for these results, really interesting stuff. And thanks also for your comprehensive VGA compatibility matrix - I've been using that for years 👍

TheMobRules wrote on 2026-01-29, 17:58:

My settings are: […]
Show full quote

My settings are:

JP1 1-2 and 3-4 closed, 5-6 open
W1 2-3 closed
W2 2-3 closed

So it seems Jp1 3-4 is the only difference, give it a try and let me know! Documentation is scarce about this card so I cannot tell if that jumper is for 0WS...

Yes, I can confirm, with JP1 2-3 closed the VLBUS3300 card is 25% faster in Wolfenstein 3D.
I have just started retesting the retesting this card with this jumper settings. Thanks again!

badmojo wrote on 2026-01-31, 01:17:

Yes I think that's the local bus timing selector jumper. I have a card based on the same chipset - different layout but the jump […]
Show full quote

Yes I think that's the local bus timing selector jumper. I have a card based on the same chipset - different layout but the jumpers all look the same:

https://theretroweb.com/expansioncards/s/west … elerator-vl-plu

And I've found the same thing - when the jumpers are set right it performs up there with the Tsengs, etc.

@Gona thanks so much for these results, really interesting stuff. And thanks also for your comprehensive VGA compatibility matrix - I've been using that for years 👍

Thanks, and you are welcome. The VGA compatibility matrix is my favorite project. 😀

Speaking of the WD90C33, tried that card on the Pentium Forex board and it also timedemos Doom faster than the Trio32 there - 35.93 FPS vs 35.48.
Upgraded the P100 to a P133 (in preparation for comparing to my Cobra board which as a Socket 4 board can only take up to P133) and it goes up to 39.33 FPS. Respectable VLB performance!

Mine is the same as pshipkov's photo here, I think the jumpers are the same as these VLBUS3300 cards?
I received mine configured as JP1 1-2 and 3-4 closed, 5-6 open, W1 1-2 and W2 2-3. Does anyone know what the W1 and W2 ones do?

Gona wrote on 2026-01-31, 11:12:

I have just started retesting the retesting this card with this jumper settings. Thanks again!

Retested. In Doom it is 5th now. https://gona.mactar.hu/VLB_benchmarks/ page updated.

Gona wrote on 2026-01-31, 19:39:

Gona wrote on 2026-01-31, 11:12:

I have just started retesting the retesting this card with this jumper settings. Thanks again!

Retested. In Doom it is 5th now. https://gona.mactar.hu/VLB_benchmarks/ page updated.

Very nice to see the WD90C33zz move up.

I still suspect that if you tested a Mach32 VLB Graphics Wonder (the dram version), I suspect it would be in the middle of the pack with the rest of the dram cards instead of the low end of the bell curve with the Oak, early trident, and the other VRAM cards.

douglar wrote on 2026-01-31, 20:03:

Gona wrote on 2026-01-31, 19:39:

Gona wrote on 2026-01-31, 11:12:

I have just started retesting the retesting this card with this jumper settings. Thanks again!

Retested. In Doom it is 5th now. https://gona.mactar.hu/VLB_benchmarks/ page updated.

Very nice to see the WD90C33zz move up.

I still suspect that if you tested a Mach32 VLB Graphics Wonder (the dram version), I suspect it would be in the middle of the pack with the rest of the dram cards instead of the low end of the bell curve with the Oak, early trident, and the other VRAM cards.

I'd like to try this sometime. My ATI collection is in a sad state; only the EXM195 is operational. My EXM229 is missing a BIOS chip, which is still better than my EXM301 which I can't even find anywhere after moving house.

Beerfloat wrote on 2026-01-31, 19:31:

I received mine configured as JP1 1-2 and 3-4 closed, 5-6 open, W1 1-2 and W2 2-3. Does anyone know what the W1 and W2 ones do?

W1 is "CPU type selection", 486 / 386
W2 is "VL BUS Selection", Enabled / Disabled

Ref: https://theretroweb.com/expansioncards/s/west … elerator-vl-plu

I don't really know what they do. W2 sounds like it should impact performance in a VLB system but it doesn't in my experience.

badmojo wrote on 2026-01-31, 22:00:

W1 is "CPU type selection", 486 / 386 W2 is "VL BUS Selection", Enabled / Disabled […]
Show full quote

W1 is "CPU type selection", 486 / 386
W2 is "VL BUS Selection", Enabled / Disabled

Ref: https://theretroweb.com/expansioncards/s/west … elerator-vl-plu

I don't really know what they do. W2 sounds like it should impact performance in a VLB system but it doesn't in my experience.

Haha the plot thickens. What could those descriptions even mean?!

badmojo wrote on 2026-01-31, 22:00:

I don't really know what they do. W2 sounds like it should impact performance in a VLB system but it doesn't in my experience.

The manual for that card provides some additional details: W2 is for connecting/disconnecting "data or code status" from the VL bus... maybe it refers to the LRDY signal? Just speculation from my side.

In any case, it seems the jumpers on that card can be matched one to one with the VLBUS3300: W1 and W2 should be the same, and then the jumper block JP1 on the VLBUS3300 would be JP2 on the "VL35".

I've been thinking about this since you posted your benchmarks. One thing to consider in VLB benchmarks in addition to whether the card uses VRAM vs DRAM is if the card has 1MB or 2MB of ram

While VRAM is going to usually take a small performance hit in DOS games, the difference between 1MB vs 2MB depends on the card.

Some cards like the Mach32 should see a performance increase going from 1MB to 2MB because the chip starts using interleaved memory access, which should reduce contention with the RAMDAC

Other cards like the S3 Trio might see a performance decrease going from 1MB to 2MB. Not sure why the 732 was faster than the 764 at DOS games, but was. If I had to guess, it could be because more chips required slower memory timing or maybe an extra wait crept in because of the extra banks memory banks.

None of these will be a big difference, maybe 3-5%, but when so many cards are clustered closely together in the charts, it can have a big impact on the ranking.

douglar wrote on 2026-02-03, 19:00:

I've been thinking about this since you posted your benchmarks. One thing to consider in VLB benchmarks in addition to whether the card uses VRAM vs DRAM is if the card has 1MB or 2MB of ram

None of these will be a big difference, maybe 3-5%, but when so many cards are clustered closely together in the charts, it can have a big impact on the ranking.

I think about the same.
Does the Tseng Labs ET4000/W32i use interleave at all on that 1 MB card or will it run faster with 2 MB?