Continued to search. Found a bunch of issues related to the board having issues with Nvidia Cards. As this was just something I wanted to use to compare performance, that's not something I will follow up anymore, and I will focus on getting decent performance out of the Voodoo 3 now.

After the 2.2 ALI driver on vogons didnt do anything (as mentioned in initial post), I tried older versions. With integrated 2.13 it installed USB, and with integrated 1.91 which I found on some Russian DOS page it installed AGP and IDE cache.
http://old-dos.ru/index.php?page=files&mode=f … =show&id=101149

Got some errors when booting in the dos window, but it seems like installation went ok.

Ran TimeDemo Benchmark again and noticed that it actually always defaults back to direct3D after game restart (glide is entered, but you dont have reflections and once you click on change, the preselected entry is direct3D again)
Average FPS increased from 30 to 37 on 512x384. Better, but still horrible.

Now Installed Ali AGP Driver 1.9, also from beforementioned russian dos page.
Did benchmark again, now at 37-38 fps. So an insignificant increase. And still bad.

Going to check impact of some BIOS switches next (but that will be in a few hours). Found some info here: Voodoo Banshee + SB16 problem!

Is this an Aladdin V board? If so, there are certain GART tweaks that have to be enabled through the ALi AGP Utility otherwise performance tanks even on 3Dfx boards.

Yeah, its an Ali V board. You got any link to a guide for that GART stuff by any chance?

Try Gigabyte GA-5AX driver, it works fine for me:
https://download1.gigabyte.com/Files/Driver/m … ff29eba94a3543e

Kordanor wrote on 2023-06-22, 12:21:

Yeah, its an Ali V board. You got any link to a guide for that GART stuff by any chance?

I'd try this.

Garrett W wrote on 2023-06-22, 12:18:

Is this an Aladdin V board? If so, there are certain GART tweaks that have to be enabled through the ALi AGP Utility otherwise performance tanks even on 3Dfx boards.

Thanks. Tested and it's unchanged. Around the same 36-37 fps as before with Ali AGP Driver 1.9.

Garrett W wrote on 2023-06-22, 13:56:

I'd try this.

Thanks, first tested:
"If possible, set GAT to 0 and Frame Buffer Size to "ALL". Also, enable Write Allocation without Hole. Once you've done these changes, try benchmarking once again."
No changes in FPS.

Then I tested:
"Write allocation enabled without holes
K6 features mask to 3
GAT to 1 (not 2)
AGP2X and sidebanding enabled
Write combining enabled"
Result: FPS are now down to 33.

Then for sience I did the force side band disable.
Result: Slightly up again to 36...

Enabled it again and enabled PCI Mode.
Result: 32 fps

Then I changed it back to have a more reasonable baseline:
PCI Off
Disable Sideband Off
AGP X2
K6 EWBC 3
GAT 2
Frame Buffer: All
Write Allocation: Enable without hole
Result: 35 fps

Then I read about the memory issue. Not sure if I am affected, but well, easy engough to take out 128MB of RAM.
Result: 41 fps

So I guess I do have the bug? The frame rate is still terrible though. So that's not it. Consider that I am testing on 512x384 and NOT on the usual 1024x768
Will test some bios settings first. Will keep the RAM out for now.

I am reverting the change after each test. Let me know if you got any recommendations. I also attach the settings I am running at as screenshot.

1. AGP Read Burst: Enabled -> Disabled
Result: 41 fps (no change)

2. ISA Line Buffer: Disabled -> Enabled
Result: 41 fps (no change)

3. Passive Release: Enabled -> Disabled
Result: 42 fps (basically no change)

4. Delay Transaction: Disabled -> Enabled
Result: 42 fps (basically no change)

5. VGA Frame Buffer: Disabled->Enabled
Result: 42 fps (basically no change)

6. Data Merge: Disabled->Enabled
Result: 42 fps (basically no change)

So basically...nothing matters.

Now I also tested with a K6 2 350Mhz CPU, resulting in 34 fps. So if there is a bug in combination with the CPU, which I dont think, it would need to apply to both.

Did a bunch of additional tests including different drivers and agp settings. The differences are really small (as long as you stick to some of the latest drivers). The biggest jump was by removing the RAM and limiting it to 128Mb
I mostly compared the results now to this post, and they seem to be on par:
Re: Asus P5A very slow AGP / Video cards
Unlike him I could not see any differences using different ram. I got tons of modules but not all have the Speed (e.g. 2-2-2) on them. Tried single sticks and pairs, but ultimately it made no difference.

Good for you, should had pay attention to cacheable range as well to rule out the performance, there is plenty of information on this chipset online to tell you what's cachable range memory to not be exceeded.

Same issue with intel chipsets all of them is 64MB cacheable except HX with second tag ram cache, 512MB.

Cheers,

pentiumspeed wrote on 2023-06-25, 17:23:

Good for you, should had pay attention to cacheable range as well to rule out the performance, there is plenty of information on this chipset online to tell you what's cachable range memory to not be exceeded.

Same issue with intel chipsets all of them is 64MB cacheable except HX with second tag ram cache, 512MB.

Cheers,

The issue is that with these items its a bit like searching for an illness on the web. You will always find people where some stuff is not working. Just telling "individual occurences" apart from general problems seems to be quite more difficult here.

True enough.

Cheers,

After (quickly) reading through this thread, it doesn't seem like you actually have a problem.

Kordanor wrote on 2023-06-21, 15:10:

While I used a couple of benchmarks, the one I got most reference to is the Unreal Benchmark
In found this video here: https://www.youtube.com/watch?v=w9CuKfQICrU
And it shows the Voodoo 3 3000 with 1024x768 on 58.5 fps

Unreal is a very CPU limited game/benchmark. You can't directly compare the performance you get on a vanilla K6-2, with the performance seen in that video (which was running on a Duron 800, a CPU that is A LOT faster than your K6-2 500).

Kordanor wrote on 2023-06-21, 15:10:

The other benchmarks I did was:
3DMark99 on 800x600 shows 161 3DMarks 5546 CPU 3DMarks
3DMark00 with standard setting shows 1048 3DMarks

You mention here getting 161 points in 3DMark 99. Is that a mistake? That score is way too low (it's as if you are hitting the 3dmark 99 "0 FPS" bug, but that's not possible with a K6-2+ CPU). Your CPU should score ~ 2500 points.
Your 3DMark00 score is a bit low, you should get ~ 1200 - 1300 points with a Voodoo 3 on a K6-2 500 (which will probably be the case if you revert to 128 MB of RAM).

bloodem wrote on 2023-06-25, 19:59:

After (quickly) reading through this thread, it doesn't seem like you actually have a problem. […]
Show full quote

After (quickly) reading through this thread, it doesn't seem like you actually have a problem.

Kordanor wrote on 2023-06-21, 15:10:

While I used a couple of benchmarks, the one I got most reference to is the Unreal Benchmark
In found this video here: https://www.youtube.com/watch?v=w9CuKfQICrU
And it shows the Voodoo 3 3000 with 1024x768 on 58.5 fps

Unreal is a very CPU limited game/benchmark. You can't directly compare the performance you get on a vanilla K6-2, with the performance seen in that video (which was running on a Duron 800, a CPU that is A LOT faster than your K6-2 500).

Kordanor wrote on 2023-06-21, 15:10:

The other benchmarks I did was:
3DMark99 on 800x600 shows 161 3DMarks 5546 CPU 3DMarks
3DMark00 with standard setting shows 1048 3DMarks

You mention here getting 161 points in 3DMark 99. Is that a mistake? That score is way too low (it's as if you are hitting the 3dmark 99 "0 FPS" bug, but that's not possible with a K6-2+ CPU). Your CPU should score ~ 2500 points.
Your 3DMark00 score is a bit low, you should get ~ 1200 - 1300 points with a Voodoo 3 on a K6-2 500 (which will probably be the case if you revert to 128 MB of RAM).

Yeah, I noticed that the Unreal Benchmark isn't really representative and I should stick to others.
Yes that 3D Benchmark was a typo. It was 1611
You can check out my benchmarks after removing the Ram and tested with various settings in the table in the bottom. I am landing right at the mentioned 2500 Mark 😀
And the 1350 on 3DMark00 seem also be fine.

Kordanor wrote on 2023-06-25, 20:05:

Yeah, I noticed that the Unreal Benchmark isn't really representative and I should stick to others. Yes that 3D Benchmark was a […]
Show full quote

Yeah, I noticed that the Unreal Benchmark isn't really representative and I should stick to others.
Yes that 3D Benchmark was a typo. It was 1611
You can check out my benchmarks after removing the Ram and tested with various settings in the table in the bottom. I am landing right at the mentioned 2500 Mark 😀
And the 1350 on 3DMark00 seem also be fine.

Oh, right! Did not see that table. 😁 So, indeed, your performance is right where it should be.

If you want an upgrade, the 5ALI61 does support K6-2/3+ CPUs (after a BIOS update), and that should give you a very nice performance boost.
Make sure you also read this thread if you haven't done so: SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod

I was playing with two Ss7 PCs recently - one with ALI AladdinV (5ALI61) and the other one with VIA MVP3 (DFI P5BV3+). The goal was to identify which chipset is better. Here are my results with amigamerlin 29 driver:

Ali Aladdin, K6-2+ 600Mhz, 128MB, Voodoo 3 3000
3D Mark 99 3456 (8049 cpu)
3D Mark 2000 1800
Quake II SW 640x480/800x600/1024x768 25.5/19.0/13.1
Quake II GL 640x480/800x600/1024x768 58,5/58,8/53.7

VIA MVP3, K6-2+ 600Mhz, 256MB, Voodoo 3 3000
3D Mark 99 3434 (7640 cpu)
3D Mark 2000 1754
Quake II SW 640x480/800x600/1024x768 24.2/18.0/12.3
Quake II GL 640x480/800x600/1024x768 57.3/57.4/53

Ali Aladdin, K6-2+ 600Mhz, 128MB
Aida 3.94 MEM 332/98
SiSoft Sandra CPU 1534/749
SiSoft Sandra MEM 172/172

VIA MVP3, K6-2+ 600Mhz, 256MB
Aida 3.94 MEM 252/119
SiSoft Sandra CPU 1532/748
SiSoft Sandra MEM 161/163

Definitely Ali is better but only marginally. I think that upgrading to K6-III/K6-2+/K6-III+ is a must for any SS7 system. Here is a reference to my P3 system with P3B-F:

440BX, PIII 450Mhz, 256MB, Voodoo 3 3000
3D Mark 99 3819 (7043)
3D Mark 2000 2405
Quake II SW 640x480/800x600/1024x768 27.6/21.1/15.3
Quake II GL 640x480/800x600/1024x768 59.3/59.9/54.2

440BX, PIII 450Mhz, 256MB
Aida 3.94 MEM 495/171
SiSoft Sandra CPU 1215/602
SiSoft Sandra MEM 262/311

Ss7 system with 600+ cpu is on par with P3 450 and both feels almost equal. Of course with faster P3 the game will change while the Ss7 system is on the limit.

Thank you! Nice tests!

drianov wrote on 2023-09-04, 21:21:

Ss7 system with 600+ cpu is on par with P3 450 and both feels almost equal.

Those numbers don't look right to me.

I had much higher framerates in Quake 2 with a Celeron 466 and a Voodoo 3 2000, at least in the lower resolutions. Benchmarks can be found in this post.

Joseph_Joestar wrote on 2023-09-04, 21:45:

drianov wrote on 2023-09-04, 21:21:

Ss7 system with 600+ cpu is on par with P3 450 and both feels almost equal.

Those numbers don't look right to me.

I had much higher framerates in Quake 2 with a Celeron 466 and a Voodoo 3 2000, at least in the lower resolutions. Benchmarks can be found in this post.

Yes, because I’m testing with default Quake 2. If I run your version (Quake2 v3.20 (with 3DFX miniGL driver v1.49)) I get 85.0/84.0/76.9 but this way I am unable to compare the result to the result of other cards / motherboards / cpus that I have.
I wrote my post just to confirm that this motherboard 5ALI61 offers decent performance with Voodoo 3 3000 AGP compared to motherboard with VIA chipset or other platform within the same range.
Actually the only thing that must be changed is BIOS over default settings is “Primary Frame Buffer” to ALL