Solved, thank you for you help 😀

Sorry to resurrect this old thread but I don’t think there was a definite resolution to the issue. I have some more information to add.

On my Gigabyte GA-5AA Rev 3.2 (running K6-2 550+) machine I have a Voodoo 2 in SLI installed. I get 48 FPS on timedemo 1 with Quake 2 (800x600). According to Phil’s Voodoo scaling project I should be getting 67.7 FPS. On 3DMark99 I get 2100. Terrible performance. I then installed the Voodoo 2 SLI into my EPoX EP-MVP3C2 (running K6-2 550+) and I got similar results.

I then moved the Voodoo 2 SLI to my AOpen AX59 Pro rev 1.31 (running K6-2 550+) and I get 98.6 FPS with Quake 2. Performance went through the roof. Way higher score than what Phil is getting. https://www.philscomputerlab.com/uploads/3/7/ … ject.pdf#page63

Still need to test 3DMark99…

zuldan wrote on 2025-02-22, 10:50:

Sorry to resurrect this old thread but I don’t think there was a definite resolution to the issue. I have some more information […]
Show full quote

Sorry to resurrect this old thread but I don’t think there was a definite resolution to the issue. I have some more information add.

On my Gigabyte GA-5AA Rev 3.2 (running K6-2 550+) machine I have a Voodoo 2 in SLI installed. I get 48 FPS on timedemo 1 with Quake 2 (800x600). According to Phil’s Voodoo scaling project I should be getting 67.7 FPS. On 3DMark99 I get 2100. Terrible performance. I then installed the Voodoo 2 SLI into my EPoX EP-MVP3C2 (running K6-2 550+) and I got the similar results.

I then moved the Voodoo 2 SLI to my AOpen AX59 Pro rev 1.31 (running K6-2 550+) and I get 98.6 FPS with Quake 2. Performance went through the roof. Way higher score than what Phil is getting. https://www.philscomputerlab.com/uploads/3/7/ … ject.pdf#page63

Still need to test 3DMark99…

Yes, I haven't found a solution for this issue, I wanted to do further tests but have had a busy couple of years. Hopefully I can investigate further during the Summer.

Yeah it is a very odd issue, I would like to find out more. Unfortunately limited time and many of my boards I would like to recap before I use them much further. I've got a few dead SS7 boards due to bad caps, although should be fixable.

It's funny, everyone always said 3dfx cards were better with K6 machines, but at least in this case the nvidia cards are better.

mwdmeyer wrote on 2025-02-22, 12:51:

Yeah it is a very odd issue, I would like to find out more. Unfortunately limited time and many of my boards I would like to recap before I use them much further. I've got a few dead SS7 boards due to bad caps, although should be fixable.

It's funny, everyone always said 3dfx cards were better with K6 machines, but at least in this case the nvidia cards are better.

Apparently Rocket202 found a solution but did not disclose it:

Rocket202 wrote on 2022-08-25, 22:00:

Solved, thank you for you help 😀

I have thought about this issue from time to time, would be nice to know what was going on, even though I solved it by changing to whole other motherboard 😏

So I got my GA-5AA Rev 1.1 out and did a quick test.

timedemo 1 (demo1.dm2) with Quake 2 (800x600)

GA-5AA Rev 1.1 - 56 FPS (Voodoo 2 Single) - Exactly what it should be according to Phil’s benchmarks
GA-5AA Rev 3.1 - 48 FPS (Voodoo 2 SLI)

So it appears the GA-5AA Rev 1.1 does not suffer from the Voodoo performance bug.

Just to recap…

SS7 boards with the Voodoo performance bug

- EPoX-ProNiX EP-MVP3C2 (MVP3)
- Gigabyte GA-5AA Rev 2.1 (ALADDiN V)
- Gigabyte GA-5AA Rev 3.1 (ALADDiN V)

SS7 boards without the Voodoo performance bug

- Gigabyte GA-5AX (ALADDiN V)
- ASUS P5A (ALADDiN V)
- Gigabyte GA-5AA Rev 1.1 (ALADDiN V)
- AOpen AX59 Pro (MVP3)

I have the GA-5AA Rev 1.1 wiith Windows 98SE

K6-III+ 450,
256MB PC100
Voodoo 3 3000 16MB.

ALi Gart driver 2.13

Here’s my 3DMartk99 results: (800x600)
I don’t know why the results are so low.

I did a lot of testing on my P3 500 with a Voodoo 3 3500 lately because of similar issues.
It seems the later drivers and directx 8/9 hurt performance.
Directx 6 and 7 provide the same scores in 3dmark99, about 4700ish. With dx8 the score drops to about 4000, dx9 even lower to 3800ish.
1.04 driver offers best performance, 1.05, 1.06 and 1.07 scores lower. Amigamerlin driver performs even worse.
Also the drivers exclusively for the 3500 seem to perform a bit worse then the general voodoo 3 drivers.
Highest 3dmark99 score for me was dx7 with 1.04 drivers. Hope I could help

I was a driver beta tester for 3Dfx before the when belly up.
I vaguely remember them saying that the last set were going to perform more poorly, but would be more compatible (?) and help the cards last longer...
I could be misremembering though...

I can add some data to this... I have a system with FIC VA-503+ (MVP3), Voodoo 3000 agp, K6+2+/570 at 616 (5.5 x 112). Here are my benchmark scores:

3DMark99 3D=3641, CPU=8058
Unreal gold 74 fps at 1024x768 in glide

I think this system does not have the bug. Unreal fps seems high compared to Phil's project for V2 sli maybe I tested it differently / wrong.

mistermister wrote on 2025-04-02, 14:08:

I can add some data to this... I have a system with FIC VA-503+ (MVP3), Voodoo 3000 agp, K6+2+/570 at 616 (5.5 x 112). Here ar […]
Show full quote

I can add some data to this... I have a system with FIC VA-503+ (MVP3), Voodoo 3000 agp, K6+2+/570 at 616 (5.5 x 112). Here are my benchmark scores:

3DMark99 3D=3641, CPU=8058
Unreal gold 74 fps at 1024x768 in glide

I think this system does not have the bug. Unreal fps seems high compared to Phil's project for V2 sli maybe I tested it differently / wrong.

I'll dig my Voodoo 3 out later and do some tests too. I have a MVP3 chipset K6-2+ running at 600MHz.

Is there any particular driver version I should test with to keep my result in line with other testers?

mistermister wrote on 2025-04-02, 14:08:

I can add some data to this... I have a system with FIC VA-503+ (MVP3), Voodoo 3000 agp, K6+2+/570 at 616 (5.5 x 112). Here ar […]
Show full quote

I can add some data to this... I have a system with FIC VA-503+ (MVP3), Voodoo 3000 agp, K6+2+/570 at 616 (5.5 x 112). Here are my benchmark scores:

3DMark99 3D=3641, CPU=8058
Unreal gold 74 fps at 1024x768 in glide

I think this system does not have the bug. Unreal fps seems high compared to Phil's project for V2 sli maybe I tested it differently / wrong.

There's a pattern here - notice how the ones unaffected have the older 586B southbridge and the affected ones have 596 southbridge?

Frankly, this is the first time I’m hearing of VIA-based boards being affected. The issue we were talking about in this thread affects later revisions of Ali Aladdin V chipsets (and, apparently, only on certain boards?!).
On the other hand, the VIA problem that @zuldan encountered is probably different, with a different root cause.

bloodem wrote on 2025-04-03, 01:15:

Frankly, this is the first time I’m hearing of VIA-based boards being affected. The issue we were talking about in this thread affects later revisions of Ali Aladdin V chipsets (and, apparently, only on certain boards?!).
On the other hand, the VIA problem that @zuldan encountered is probably different, with a different root cause.

My GA-5AA version 1.1 has the revision E chipset.

Is my score in my post above at 450MHz normal or low?

I have an Asus P5A-B and Chaintech 5AGM2 that I’ll test with a K6-2+ 550 as soon as I have time.

byte_76 wrote on 2025-04-03, 06:09:

My GA-5AA version 1.1 has the revision E chipset.

Is my score in my post above at 450MHz normal or low?

I have an Asus P5A-B and Chaintech 5AGM2 that I’ll test with a K6-2+ 550 as soon as I have time.

I don't think I ever tested a K6-3+ @ 450MHz, but here's a video I created some years ago, showcasing the 3DMark99 performance (among others) with the CPU running at 633 MHz (FSB 105) & a Voodoo 3 3000 @ 3500 clocks.
I would say that you are well below what I would expect to see, even for a lower CPU frequency such as 450MHz.

bloodem wrote on 2025-04-03, 01:15:

Frankly, this is the first time I’m hearing of VIA-based boards being affected. The issue we were talking about in this thread affects later revisions of Ali Aladdin V chipsets (and, apparently, only on certain boards?!).
On the other hand, the VIA problem that @zuldan encountered is probably different, with a different root cause.

Same goes for VIA - boards with the old 586 southbridge (e.g DFI P5BV3+) are not affected, only the ones with 596 southbridges (e.g DFI K6BV3+/66) have the Voodoo bug (and I can confirm my own K6BV3 being one of them).

This might also extend to Pentium II and III boards using the 596 southbridges as well, however I've yet to test that theory.

I have done some BIOS tuning and set the clock multiplier to 5.5 for a processor speed of 550Mhz.

Here are the new bench results:

PcBytes wrote on 2025-04-03, 13:50:

Same goes for VIA - boards with the old 586 southbridge (e.g DFI P5BV3+) are not affected, only the ones with 596 southbridges (e.g DFI K6BV3+/66) have the Voodoo bug (and I can confirm my own K6BV3 being one of them).

This might also extend to Pentium II and III boards using the 596 southbridges as well, however I've yet to test that theory.

That just sounds like a coincidence. The southbridge should be irrelevant, especially when it comes to AGP cards. It mostly just handles peripherals and definitely not AGP signaling.
I also have a newer DFI K6XV3+/66 board (which is the ATX version of your K6BV3+/66 board), this one also has the 596B south bridge, and it works just fine with Voodoo 3 AGP cards (most likely also with Voodoo 3 PCI cards, but I never bothered testing any on it). This particular board was also featured in one of Phil's videos, and performance is right where you would expect it to be (3332 points in 3DMark 99, with a Voodoo 3 3500): https://www.youtube.com/watch?v=ZmQnEdoZyh8

Just as a reminder for all those who encounter lower than expected performance - make sure that you are NOT using any peripherals such as USB 2.0 adapters, those can kill 30% of your performance right off the bat on certain boards.

I’m going to update my list above. I just received my Banshee today and put it in the EPoX-ProNiX EP-MVP3C2 (MVP3) running a K6-2+ 570 @ 600 (100 x 6) with v1.04.00 drivers. Look at the result…

No performance issues with a Banshee (PCI) on MVP3. What’s strange is my Voodoo 3 (AGP) was getting a score of 2400 on this same motherboard? I had a refresh install of Win98 SE with the chipset drivers installed (including the AGP driver).