asdf53 wrote on 2026-05-26, 18:57:

It's still not normal that your 3DMark99 score is lower than your 3DMark00 score. Could it be that you ran it with Vsync enabled?

It seems you were correct! Many thanks! 3DMark 99 score went up from 7329/20272 up to 9885/20403.

3DMark 2000 unchanged though, so that obviously does disable vsync itself.

asdf53 wrote on 2026-05-25, 19:07:

memory everest 2.20: 984 378 178ns (read/write/access)
memory aida bios 2 bank interleave: 995 1025 144
memory aida bios 4 bank interleave: 995 1027 143

Everest I get 961 422 159ns

Looking at the other systems in the list, it seems my scores are about right

Aida 64 7.6 just crashes

asdf53 wrote on 2026-05-25, 19:07:

final reality agp test: 237.76

Final Reality AGP test gives 247.53 images per second with 16Mb default setting. If that's what you did, then my result seems about right.

With 36Mb (the max) it scores 142.99. If you did that, then my score is very low.

Based on the above I am thinking that now I have improved memory timings, and disabled vsync, that the system is performing well enough. I just had incorrect expectations because Phil probably tested the card on a much more powerful platform. Nothing wrong with that, but it's not period correct and I didnt know the card was so severely bottlenecked by the platforms of the day.

My quake 3 benchmark at 1024x768 16 bit is now 137.3fps (153.3 without sound). Deus Ex gets about 30FPS on the starting level at 1024x768 16bit. That's probably the most system taxing game I am likely to run on this.

Make sure your AGP is running in actual AGP mode (GART driver), this can be checked via PowerStrip utility (if AIDA64 fails to load).
Note : AIDA64 only works up to v7.70.7500 on 32-bit systems (and even that can be limited to .zip only, depends on actual OS).

3DMark 99MAX :

3DMark 2k :

Quake 3 Arena, demo001 @ 1024x768 32-bit color

I can only provide 3DMark 2001SE scores for now (since my benchtable isn't connected at the moment, although still assembled), but here are some results with a *very* strong system that bottlenecks none of these cards:

download/file.php?mode=view&id=243002

With a Athlon XP 2500 seen as a XP 1500 (1311 MHz 105x12.5) on a K7T Pro2 mainboard and a GeForce 4 MX440, I get 7023 in 3DMark99, and 7958 with 3DMark2000.
Not sure about the driver, probably 45.23.
So basically same as you with a quite equivalent hardware.

NorsteinB wrote on 2026-05-27, 16:04:

I can only provide 3DMark 2001SE scores for now (since my benchtable isn't connected at the moment, although still assembled), but here are some results with a *very* strong system that bottlenecks none of these cards:

download/file.php?mode=view&id=243002

Many thanks. I score 5115 compared to your score of 7667. Quite a difference - 50% more! Though I think your CPU is certainly a big part of that increase.

nali wrote on 2026-05-28, 02:54:

With a Athlon XP 2500 seen as a XP 1500 (1311 MHz 105x12.5) on a K7T Pro2 mainboard and a GeForce 4 MX440, I get 7023 in 3DMark99, and 7958 with 3DMark2000.
Not sure about the driver, probably 45.23.
So basically same as you with a quite equivalent hardware.

Thanks. That's is reassuring to know. Whilst I know that the GPU is capable of more, it does seem that it is performing as expected in the system that its in. I am quite happy with it as a period-correct build now.

I basically want this as a software development system for learning DOS programming up to directx 7. For this, it seems perfect.

It's secondary role is for playing any games of the era. I will make a Geforce 4600-based system for Directx 8 games, or to play more demanding directx 7 games.

agent_x007 wrote on 2026-05-27, 15:18:

Make sure your AGP is running in actual AGP mode (GART driver), this can be checked via PowerStrip utility (if AIDA64 fails to l […]
Show full quote

Make sure your AGP is running in actual AGP mode (GART driver), this can be checked via PowerStrip utility (if AIDA64 fails to load).
Note : AIDA64 only works up to v7.70.7500 on 32-bit systems (and even that can be limited to .zip only, depends on actual OS).

3DMark 99MAX :

The attachment 3DMark99 MAX.PNG is no longer available

3DMark 2k :

The attachment 3DMark2k.PNG is no longer available

Quake 3 Arena, demo001 @ 1024x768 32-bit color

The attachment Quake 3 Arena Demo001 1024x768x32.PNG is no longer available

Wow. Your 3DMark 99 absolutely blows my system out of the water. That's crazy!

Anyway, this is what my Powerstrip says:

Those GF 4 MX are basically GF2, so they are quite nice for old systems.
The Nvidia driver for GF2 even works for BeOS 4.5, and of course BeOS 5 😀
Same for QNX 6.2.1 and OS2.
So if 3D perf are not the main goal, it's a quite good card.
Anyway perf are not so bad for Win98 area games. I get 85 fps in Unreal 1 in 1024x768, 330 fps in Quake 1, 60 for Half Life Blue Shift, so I'm happy with this 😀

nali wrote on 2026-05-28, 08:06:

Those GF 4 MX are basically GF2, so they are quite nice for old systems. The Nvidia driver for GF2 even works for BeOS 4.5, and […]
Show full quote

Those GF 4 MX are basically GF2, so they are quite nice for old systems.
The Nvidia driver for GF2 even works for BeOS 4.5, and of course BeOS 5 😀
Same for QNX 6.2.1 and OS2.
So if 3D perf are not the main goal, it's a quite good card.
Anyway perf are not so bad for Win98 area games. I get 85 fps in Unreal 1 in 1024x768, 330 fps in Quake 1, 60 for Half Life Blue Shift, so I'm happy with this 😀

Yes, agreed. Most games I tried are running well. Deus Ex (from year 2000) was probably the directX 7 game that it struggled most with. But this game is well known for this. The quoted min specs for that game are complete nonsense. Though I believe that game ran better on Glide than Direct3D. I haven't tried it on Glide yet.

NorsteinB wrote on 2026-05-27, 16:04:

I can only provide 3DMark 2001SE scores for now (since my benchtable isn't connected at the moment, although still assembled), but here are some results with a *very* strong system that bottlenecks none of these cards:

MX460 (stock) :

MX440 (OC'ed) :

EDIT : GF2 Ti after OC with better VRAM chips 😉

agent_x007 wrote on 2026-05-28, 17:54:

MX460 (stock) : […]
Show full quote

MX460 (stock) :

The attachment 3DMark 01SE.PNG is no longer available

MX440 (OC'ed) :

The attachment 3DMark 01SE.PNG is no longer available

EDIT : GF2 Ti after OC with better VRAM chips 😉

The attachment 3DMark 01SE OC2.PNG is no longer available

So does this mean that even NorsteinB's X2 555 CPU was bottlenecking the GPUs? Apart from the OC MX 440, your scores are quite a bit higher.

In 800x600 I get 44fps with the 440 MX, and 30 fps with a Voodoo2.

RetroPCCupboard wrote on 2026-05-28, 18:45:

So does this mean that even NorsteinB's X2 555 CPU was bottlenecking the GPUs? Apart from the OC MX 440, your scores are quite a bit higher.

Personally, I think it's all in the drivers (we don't know which ones were used by him).
However his system should be better suited for this type of tests (since it has IMC).

nali wrote on 2026-05-28, 20:52:

In 800x600 I get 44fps with the 440 MX, and 30 fps with a Voodoo2.

Interesting. I was thinking of comparing with Voodoo 3 or Voodoo 2 SLI to see how that does.

I have only 1 Voodoo2, and I got it for cheap 😀 20$ at a local store with a bit of old stock.
I remember trying Deus Ex back in the days with a Voodoo3 3000, and was not impressed by the perfs.
I think it was the end of the era for Voodoo3, it needs more power than Unreal 1 or UT.
Even Undying or Wheel of Time needed more power.
But my PC was a dual Celeron at 490 MHz (BP6), so a more powerful cpu could help.

RetroPCCupboard wrote on 2026-05-25, 06:34:

Hi All, […]
Show full quote

Hi All,

I am currently testing a Geforce 4 MX 460 with a Thunderbird 1.4Ghz on a KT7A-RAID motherboard. Using nVidia driver 45.23

Performance I am getting is:

3DMark 99 = 7264 GPU, 19668 CPU
3DMark 2000 = 7761

I was expecting somewhere around 11-12000 on 3dMark 2000. I
Do you think the CPU is bottlenecking it, or could something else be wrong?

Which version of Via 4 in 1 are you using?

For me seems strange values, I remember on the times of these chipsets we getting low framerates due to this.

agent_x007 wrote on 2026-05-28, 21:03:

RetroPCCupboard wrote on 2026-05-28, 18:45:

So does this mean that even NorsteinB's X2 555 CPU was bottlenecking the GPUs? Apart from the OC MX 440, your scores are quite a bit higher.

Personally, I think it's all in the drivers (we don't know which ones were used by him).
However his system should be better suited for this type of tests (since it has IMC).

I used a relatively new driver IIRC, since I wanted to test all of them with the same version. Sometimes, that means you get more driver overhead compared to older versions, so an older driver might perform better with older cards.
I can't check which version it was since the system is dissambled rn, but it must have been at least 61.76.

fcm wrote on 2026-06-01, 01:28:

Which version of Via 4 in 1 are you using?

For me seems strange values, I remember on the times of these chipsets we getting low framerates due to this.

I have tried version 4.35 and 4.56. 4 35 have very slightly higher score, but within margin of error.

Interestingly I have now tried something that I didn't think would work. I installed driver 23.11, and forced the driver installation saying it is a Geforce 2 Ultra. Surprisingly, it seemed to install ok. My 3DMark 2000 score is now over 9000! Quake 2 though complains that there is no openGL.