Well... have to say, I really don't understand what's happening with Quake 3.

I don't have a Coppermine 550MHz, but I had the next best thing, a Pentium 3 Katmai 450 MHz, which I overclocked to 504 MHz (FSB 112) on an Amptron 440BX motherboard. I also threw in a GeForce 2 Ti in the mix.

I tested two Quake 3 versions:
- the one from an original CD that I have (v1.11) which has two demos: demo001 & demo002
- an ISO image that I have (v1.30) which has one demo: four.dm_66

I used a pre-existing Win98 installation on a 32 GB CF card (which had previously been used for a Core 2 Duo/i865, SS7 VIA MVP3 motherboard and... others).
I simply installed the latest 440BX chipset drivers and then sequentially tested the following nVIDIA drivers with the GeForce 2 Ti: 30.82 / 12.41 / 7.76

All tests were done at 640 x 480, with two different settings:
1. Details A: Color depth 16 bit / Lighting Lightmap / Geometric detail High / Texture Detail step 3 out of 4 / Texture quality 16 bit / Texture filter Bilinear)
2. Details B: Color depth 16 bit / Lighting Vertex / Geometric detail Low / Texture Detail step 1 out of 4 / Texture quality 16 bit / Texture filter Bilinear)

The results are very weird...
nVIDIA driver 30.82:
Quake 3 v1.11 demo001, Details A: 81.3 FPS
Quake 3 v1.11 demo001, Details B: 96.2 FPS
Quake 3 v1.30 four.dm_66, Details A: 80.4 FPS
Quake 3 v1.30 four.dm_66, Details B: 93.5 FPS

nVIDIA driver 12.41 (forced install - GeForce 2 Ti not officially supported):
Quake 3 v1.11 demo001, Details A: 81.9 FPS
Quake 3 v1.11 demo001, Details B: 95.7 FPS
Quake 3 v1.30 four.dm_66, Details A: 78.6 FPS
Quake 3 v1.30 four.dm_66, Details B: 91.8 FPS

nVIDIA driver 7.76 (forced install - GeForce 2 Ti not officially supported):
Quake 3 v1.11 demo001, Details A: 83.3 FPS
Quake 3 v1.11 demo001, Details B: 97.1 FPS
Quake 3 v1.30 four.dm_66, Details A: 81.6 FPS
Quake 3 v1.30 four.dm_66, Details B: 92.4 FPS

I don't understand... these results are better than what I got on my Athlon 750 MHz/ P3 750 MHz builds. I'll do a bit more digging to try and understand what's happening...
All in all, on an old CPU like the Katmai, even though there are some differences between driver versions / Quake 3 versions / demos, we're talking about a <5% variance, so not something worth considering.

EDIT:
Tested a bit without overclock (Katmai 450 MHz / FSB 100), driver 7.76 - still impressive results, IMO:
Quake 3 v1.11 demo001, Details A: 74.1 FPS
Quake 3 v1.11 demo001, Details B: 84.8 FPS
Quake 3 v1.30 four.dm_66, Details A: 73.4 FPS
Quake 3 v1.30 four.dm_66, Details B: 83.3 FPS

OK, mystery partly solved: the results above were without a sound card. As soon as I threw in a Yamaha YMF724 (which actually has DirectSound hardware acceleration), the FPS for Details A dropped from 80+ down to 65. That's a huge drop in performance (25%)! I'm pretty sure Anand's tests were also carried out without a sound card.
What sound card are you using, RetroSpector78? If it's one without hardware acceleration... this might (at least partially) explain your results.

EDIT: indeed, Anand does not mention anything about using a sound card in his test specs. So looking at Anand's results, he got 101 on the Coppermine 550MHz (and he definitely used low details, since he mentioned this in other subsequent reviews), while I got 97 FPS on the overclocked Katmai.

You should never run "timedemo" like benchmarks with sound, it will just bottleneck your system once you get high FPS.

Ideally the sound would still run at a fixed rate (1x speed) while the graphics subsystem could run full speed but that is not how these games are coded.

Not sure what you mean by that. Sound adds some overhead, particularly on slower CPUs - that's perfectly normal (Quake 3 sound seems to add quite a bit of overhead).
However, I've never had issues with High FPS and timedemos with sound (even on a Core i5 at 1000 FPS). It always scales accordingly with or without sound.

I always had s_mixahead tightened back in the Q3 days, the sound processing was significant.

bloodem wrote on 2021-02-16, 12:47:

OK, mystery partly solved: the results above were without a sound card. As soon as I threw in a Yamaha YMF724 (which actually has DirectSound hardware acceleration), the FPS for Details A dropped from 80+ down to 65. That's a huge drop in performance (25%)! I'm pretty sure Anand's tests were also carried out without a sound card.
What sound card are you using, RetroSpector78? If it's one without hardware acceleration... this might (at least partially) explain your results.

EDIT: indeed, Anand does not mention anything about using a sound card in his test specs. So looking at Anand's results, he got 101 on the Coppermine 550MHz (and he definitely used low details, since he mentioned this in other subsequent reviews), while I got 97 FPS on the overclocked Katmai.

In Quake 3 I did all my runs with the following

1s_initsound 0
2snd_restart
3timedemo 1
4demo dem001

That should be sufficient to turn off sound and it does indeed increase the fps (but not by a lot ... around 10-15%)

But you numbers are impressive ... even when tweaking some bios settings I can't get mine to go beyond 65fps.
The Geforce2MX and Geforce4MX do require some newer drivers. Perhaps that combined with the Via chipset as opposed to the Intel chipset makes a difference.....

Are you using any peripherals such as a PCI USB adapter, maybe? USB keyboard, USB mouse?
To debug further, I would remove everything and only leave the motherboard, CPU, RAM and VGA + PS/2 mouse & keyboard.

The GeForce 2 MX does not require newer drivers. It works fine with driver version 7.76.

Quake 3 is vey sensible to cache-speed. That's why not even my PIII 600 (Katmai) is able to beat my k6-III+@616 in Quake 3. Although normally a PII 450 is performing similar to the overclocked K6-III.
Do you have a p3 Coppermine for comparison?

Not sure about a Coppermine, but your Katmai results are on the low side. Is that with or without sound?
EDIT: Never mind, just saw that you tested with Voodoo 2 SLI. I guess the results are normal (Quake 3 benefits from cards with Hardware T&L).
Out of curiosity, I will try later today and see what results I get on my Katmai 504 MHz with a PCI Voodoo 3 2000 (which should be more or less equivalent to a Voodoo 2 SLI).

bloodem wrote on 2021-02-16, 20:23:

Are you using any peripherals such as a PCI USB adapter, maybe? USB keyboard, USB mouse?
To debug further, I would remove everything and only leave the motherboard, CPU, RAM and VGA + PS/2 mouse & keyboard.

The GeForce 2 MX does not require newer drivers. It works fine with driver version 7.76.

Removed the soundcard and disabled the USB adapter. So no USB devices in play
Q3 demo runs with initsound 0 and without a soundcard, so no bottleneck there.

Will double-check the video drivers but find it strange that a lot of cards in this system seem to max out around 60-65fps in Q3 640x480x16
Perhaps it's the Via chipset ?

bloodem wrote on 2021-02-17, 09:53:

Not sure about a Coppermine, but your Katmai results are on the low side. Is that with or without sound?

No sound. T&L cards make a huge difference indeed. But my point was not to compare with other results using a T&L-cards, but to show how much Quake 3 depends on a fast L2-Cache.

-Quake 3 demo Demo1: no sound, maxed settings apart from textures (only second highest)
-P3 600 Katmai on bx440 MSI 6119 @100mhz fsb
-K6-III+500@616 on Soyo 5EMA+ @5.5x112mhz
-V2 sli 12mb using the FastVoodo2 driver 4.6

Williwinner wrote on 2021-02-17, 10:35:

No sound. T&L cards make a huge difference indeed. But my point was not to compare with other results using a T&L-cards, but to show how much Quake 3 depends on a fast L2-Cache.

I understand that, but if for some unforeseen reason your Q3 results on the Katmai would be too low, the conclusion would be wrong. 😀
If I'm comparing a Ferrari with a Renault, but the Ferrari is running on only one cylinder... the fact that the Renault will turn out to be faster is not really relevant... you kow what I mean? 😁

I just tested Quake 3 (v1.11 / demo001) on the Katmai 504 MHz + Voodoo 3 2000 PCI:
Resolution: 640 x 480 / Color depth: 16 bit / Lighting: Lightmap / Geometric detail: High / Texture Detail: step 3 out of 4 / Texture quality: 16 bit / Texture filter: Bilinear
I got 63.8 FPS (no sound card)
EDIT: On Quake 3 v1.30 / map four.dm_66 (same settings as above): 62.1 FPS

bloodem wrote on 2021-02-17, 09:53:

Out of curiosity, I will try later today and see what results I get on my Katmai 504 MHz with a PCI Voodoo 3 2000 (which should be more or less equivalent to a Voodoo 2 SLI).

Looking forward to that. In the meantime i have some results for the k6-iii+ using a Geforce ddr for the T&L-Boost as well.
Still no sound, everything maxed except for textures @ second highest.
-TNT2 Ultra Guillemont 175/183 using Det 3.68
-V3 3500 final referance driver
-GF 256 ddr 32mb Asus using Det 8.05

bloodem wrote on 2021-02-17, 10:52:

I just tested Quake 3 (v1.11 / demo001) on the Katmai 504 MHz + Voodoo 3 2000 PCI: Resolution: 640 x 480 / Color depth: 16 bit / […]
Show full quote

I just tested Quake 3 (v1.11 / demo001) on the Katmai 504 MHz + Voodoo 3 2000 PCI:
Resolution: 640 x 480 / Color depth: 16 bit / Lighting: Lightmap / Geometric detail: High / Texture Detail: step 3 out of 4 / Texture quality: 16 bit / Texture filter: Bilinear
I got 63.8 FPS (no sound card)
EDIT: On Quake 3 v1.30 / map four.dm_66 (same settings as above): 62.1 FPS

Cool. Since i used a fresh Windows 98. I can't really think of any other differencees to my 58fps on the P3 600 apart from:
- i am using the Q3 demo version
- my bx440 board could just be quite slow and...
- ... it is a very basic one without much to tune in the bios.
- 112mhz fsb gives your Katmai a slight adavantage as well

Williwinner wrote on 2021-02-17, 10:59:

Looking forward to that. In the meantime i have some results for the k6-iii+ using a Geforce ddr for the T&L-Boost as well. Stil […]
Show full quote

Looking forward to that. In the meantime i have some results for the k6-iii+ using a Geforce ddr for the T&L-Boost as well.
Still no sound, everything maxed except for textures @ second highest.
-TNT2 Ultra Guillemont 175/183 using Det 3.68
-V3 3500 final referance driver
-GF 256 ddr 32mb Asus using Det 8.05

Those are very nice results!

Williwinner wrote on 2021-02-17, 11:13:

Cool. Since i used a fresh Windows 98. I can't really think of any other differencees to my 58fps on the P3 600 apart from: - i […]
Show full quote

Cool. Since i used a fresh Windows 98. I can't really think of any other differencees to my 58fps on the P3 600 apart from:
- i am using the Q3 demo version
- my bx440 board could just be quite slow and...
- ... it is a very basic one without much to tune in the bios.
- 112mhz fsb gives your Katmai a slight adavantage as well

Yes, to tell you the truth I'm also surprised by the performance I'm seeing with this old Katmai (on a rather cheap Amptron 440BX board).
For sure the overclocked FSB adds a substantial gain, but even when testing with no overclock results seem to be very good.

Anyway, enough with the Katmai, back to my Pentium MMX @ 290 MHz 😀

EDIT: One interesting remark: The Voodoo 3 2000 PCI severely bottlenecks the Katmai 504 MHz in quite a few games (GLQuake, Quake 2) at 640 x 480 (even with help from the Glide API)
On the other hand, other games like NFS High Stakes or Unreal get a huge boost with the Voodoo 3 when using the Glide API. NFS High Stakes is infamous for its subpar performance in Direct3D even much more powerful GPUs and CPUs, so the fact that it can run so well on a Katmai when you pair it with a Voodoo 3 is mind-blowing (hence why people love the Voodoo cards so much) 😀

RetroSpector78 wrote on 2021-02-16, 08:55:

Some results : […]
Show full quote

Some results :

• 640x480x16 low detail (textures & geometry) with Vertex lighting : 73fps (DRAM : turbo)
• 640x480x16 low detail (textures & geometry) with Lightmap lighting : 66fps (DRAM : turbo)
• 640x480x16 high detail (textures & geometry) with Lightmap lighting : 63fps. (DRAM : turbo)
• 640x480x16 high detail (textures & geometry) with Lightmap lighting : 61fps (DRAM : normal)
• 640x480x16 high detail (textures & geometry) with Lightmap lighting : 59fps (DRAM : setup defaults were 10ns != Normal)

Vertex lighting is definitely cheating 😀

With a Radeon 9800 Pro running at the completely CPU-limited resolution of 1024x768x32 w/ high detail, I was getting 31.4 fps with a P2-300, and 110.5 fps with a P3-850. Both on the same 440BX board with PC100 (2-2-2-6 timings). Your ~65 fps is right in the middle of that, which sounds about right for a 550MHz Katmai.