Those are some outrageous Quake scores for the Geforce 2MX. Does the driver support 3dnow or something?

douglar wrote on 2024-03-17, 21:23:

Those are some outrageous Quake scores for the Geforce 2MX.

I was suprised as well, but chucked it up to the motherboard being a great performer (witch it is). However, I'm not fond of running nvidia cards on super 7 PCs. Sure, the max FPS is fantastic, but there is stuttering in some games. Homeworld for example - whenever zooming in on one ship the game stutters. When zooming out in large fleet battles, more stuttering. In Half Life, again, stuttering, but with the Savage 4 or a Kyro there is no stuttering whatsoever.

douglar wrote on 2024-03-17, 21:23:

Does the driver support 3dnow or something?

I don't really know, but it's the driver version I've been using with super 7 builds for years. If it does, I'd be great if someone more knoladgeble could confirm it.

If you haven't already done so, try the 3DNow! patch for Quake 2.

It should improve the scores a bit further, especially for the Voodoo 2.

Joseph_Joestar wrote on 2024-03-17, 21:30:

If you haven't already done so, try the 3DNow! patch for Quake 2.

It should improve the scores a bit further, especially for the Voodoo 2.

I don't have the 3DNow version on this machine, but it would probably be a good idea to get it, since I've narrowed the video card choice to either the SiS 305 (low avarage FPS but very smooth) or maybe the S3 Savage 4, and every little bit of extra performance helps. I plan to test a PowerVR Kyro (1) and a Radeon DDR on the platform as well, then make a final choice. I might even expand the benchmark series to include the Riva 128ZX, the Kyro I, afformentioned Radeon DDR, the Rage 128 PRO and a Voodoo 3 3000.

IIRC, the 3DNow! patch will offer a slight boost to any card, but it's particularly effective on a Voodoo 2.

I tried it on a Voodoo 1 and a Voodoo 3, and the performance did improve, but the difference wasn't quite stellar. The 3DNow! + 3DFX renderer which uses 3dfxglam.dll is custom tailored to the K6-2 paired with a Voodoo 2, and should provide significant gains on that setup.

The reason Quake is so much faster on GeForce is the HW T&L, which Quake can make use of and is particularly impressive on a slow CPU such as the K6 series.

Garrett W wrote on 2024-03-18, 00:54:

The reason Quake is so much faster on GeForce is the HW T&L, which Quake can make use of

GLQuake? I was under the impression all the math is done by quake code, miniGL receives precalculated triangles and there is nothing for TnL to do.

No miniGL on GeForce. GLQuake is using OpenGL for which GeForce provides a full ICD. The extensions expose this feature!

I'd sooner think it's about texture memory. Note the T&L-less Savage4 also beats the V2 at both by a good margin. The margin would be even larger on some thrashy maps (like q3dm11 the best map to make a v2 sli elitist cry on). Q3 benchmarking usually takes place on Q3DM6 which is the fastest running map in the game for being enclosed and not having much texture variety or portals.

GLQuake can still give a Voodoo2 a hard time over dynamic lightmap updates causing little upload hitches.

The thing that's actually surprising me about the MX400 is that it hasn't got the same lead in everything else that it's got in the quakes. Then also I get the impression in other reviews that a Savage 4 can't actually stick it's neck out ahead of a TNT2 M64, but here it is, not a particularly fast example is it? That is taking a convincing lead over it in most things.

BitWrangler wrote on 2024-03-18, 03:26:

The thing that's actually surprising me about the MX400 is that it hasn't got the same lead in everything else that it's got in the quakes. Then also I get the impression in other reviews that a Savage 4 can't actually stick it's neck out ahead of a TNT2 M64, but here it is, not a particularly fast example is it? That is taking a convincing lead over it in most things.

It's your garden variety Savage 4 (Fastware AG460D) running at 110/121MHz. I have higher clocked versions of this card. I think it's taking a lead over most things because it's driver has less CPU overhead then say the nvidia cards. On a faster machine the GF2 MX should outperform it consistently.

I'm tempted to repeat the tests on a 2600+ Thoroughbread (2083MHz) / KT333.

Socket3 wrote on 2024-03-18, 06:19:

BitWrangler wrote on 2024-03-18, 03:26:

The thing that's actually surprising me about the MX400 is that it hasn't got the same lead in everything else that it's got in the quakes. Then also I get the impression in other reviews that a Savage 4 can't actually stick it's neck out ahead of a TNT2 M64, but here it is, not a particularly fast example is it? That is taking a convincing lead over it in most things.

It's your garden variety Savage 4 (Fastware AG460D) running at 110/121MHz. I have higher clocked versions of this card. I think it's taking a lead over most things because it's driver has less CPU overhead then say the nvidia cards. On a faster machine the GF2 MX should outperform it consistently.

I'm tempted to repeat the tests on a 2600+ Thoroughbread (2083MHz) / KT333.

I would like to see that! I was really surprised by the numbers you got with the Savage 4. I have done a fair amount of tinkering with cards of this level on my Pentium III 850Mhz 440BX test rig and while the Savage 4 was certainly among the best cards that wasn't from 3dfx or Nvidia, it never seemed to be as good as any Voodoo or TNT cards I've run. That could all just be my perception though... I haven't done formal testing exactly.

What driver versions were you using for the TNT, M64 and Voodoo 2?

Also, nice job! These numbers are very handy for people looking to put together a suitable retro PC.

... thinking on it now though, it would be cool if there was a way to use modern benchmarking tools and techniques on retro PCs. Over the past 15 years or so a lot of benchmarking sites (starting with TechReport) have started focusing more on frame times and consistency, which can paint a very different picture of smoothness and responsiveness compared to an average FPS number. I would be curious to know where cards from this era stack up in that regard (I feel like 3dfx cards in Glide would shine there), but I don't think the tools exist to measure frametimes on such old hardware\software. Actually, much more recently some reviewers (like Gamers Nexus, with the help of software being developed by Intel) are breaking benchmarking down even further to see where exactly during the pipeline the bottleneck is so that you can actually see and chart the impact of driver overhead and CPU speed on GPU render times... really amazing stuff that would have fundamentally changed the development of GPUs, drivers and APIs if it had been available 20 years ago.

... but I digress. FPS numbers are probably fine when we're dealing with frame rates mostly under 40. 😁

Ozzuneoj wrote on 2024-03-18, 06:34:

... thinking on it now though, it would be cool if there was a way to use modern benchmarking tools and techniques on retro PCs. Over the past 15 years or so a lot of benchmarking sites (starting with TechReport) have started focusing more on frame times and consistency, which can paint a very different picture of smoothness and responsiveness compared to an average FPS number.

Yeah, average FPS numbers don't always tell the complete story. For example, in Quake 3, I get 120+ FPS on a GeForce FX 5900XT at 1600x1200 using the timedemo benchmark. And yet, during actual gameplay, the FPS tanks whenever I approach a portal or a mirror, like the ones in Q3DM0 (the very first tutorial map).

Best you can do, without specialized tools, is to measure the lowest, average and highest FPS in games which allow that. For example, Unreal and UT99 can do this using their built-in benchmarks. If your lowest FPS score is above 60 while benchmarking, you can be reasonably sure that you're in for a smooth ride during actual gameplay.

Joseph_Joestar wrote on 2024-03-18, 06:52:

Ozzuneoj wrote on 2024-03-18, 06:34:

... thinking on it now though, it would be cool if there was a way to use modern benchmarking tools and techniques on retro PCs. Over the past 15 years or so a lot of benchmarking sites (starting with TechReport) have started focusing more on frame times and consistency, which can paint a very different picture of smoothness and responsiveness compared to an average FPS number.

Yeah, average FPS numbers don't always tell the complete story. For example, in Quake 3, I get 120+ FPS on a GeForce FX 5900XT at 1600x1200 using the timedemo benchmark. And yet, during actual gameplay, the FPS tanks whenever I approach a portal or a mirror, like the ones in Q3DM0 (the very first tutorial map).

Best you can do, without specialized tools, is to measure the lowest, average and highest FPS in games which allow that. For example, Unreal and UT99 can do this using their built-in benchmarks. If your lowest FPS score is above 60 while benchmarking, you can be reasonably sure that you're in for a smooth ride during actual gameplay.

Yeah, generally if minimums are at a tolerable level it will be okay. The issue really comes in those situations where performance is fluctuating a lot per second... for example, frames taking 33ms half the time and the other half taking 8ms, you'll feel pretty nasty stutters (like 120 to 30 fps) even though the frame rate would likely be recorded as 60fps by most software.

It's funny we have the luxury of even thinking about such things now. Someone trying to game on a Trident Blade3D was just happy to be able to play a game with hardware acceleration at 25fps back in the day. And if you had bags of money laying around you could get some crazy Quantum3D SLI monstrosity to hit 6ofps at 1024x768 in some games... and two years later it was worthless for any kind of 3D gaming, and your CPU was probably less than half the speed of whatever the latest thing was. 🤣

Things were so much more "interesting" back then...

It's an AGP 2x motherboard.

WIth AGP 4x, the TNT2-M64 should be +40% faster, and the Geforce 2 MX-400 should be +200-400% faster.

Thanks a lot for these interesting benchmarks. I love that you only tested budget cards, to me SS7 is a budget platform so using these feels very appropriate. I wouldn't have thought that the Savage4 performs so well and even beats a GF2 MX, especially at these low clock rates. It's also very overclockable, I remember clocking mine to around 150/160 and it didn't even get particularly hot.

The Kyro also performs very well on SS7 systems, I got 3700 points in 3dmark99 with a Kyro 2 (3380 with a GF2 MX). The Geforce also needs old drivers to perform well, some newer games are unplayable with those. The Kyro performs well across the board using the newest drivers, even in OpenGL. I think it was the best performing SS7 card out of all that I tested.

If you feel that your MVP3 board is slow, may I suggest trying these tweaks: https://www.k6plus.com/phpBB3/viewtopic.php?t=1406

On my system, this improved 3DMark99 scores on a Geforce 2MX from 3000 to 3400 points, so it's well worth the effort.

Ozzuneoj wrote on 2024-03-18, 06:34:

What driver versions were you using for the TNT, M64 and Voodoo 2?

Driver versions, sistem specs and software used are posted in the first thread, as well as where I listed the benchmark results in text format.

Ozzuneoj wrote on 2024-03-18, 06:34:

... thinking on it now though, it would be cool if there was a way to use modern benchmarking tools and techniques on retro PCs. Over the past 15 years or so a lot of benchmarking sites (starting with TechReport) have started focusing more on frame times and consistency, which can paint a very different picture of smoothness and responsiveness compared to an average FPS number. I would be curious to know where cards from this era stack up in that regard (I feel like 3dfx cards in Glide would shine there), but I don't think the tools exist to measure frametimes on such old hardware\software. Actually, much more recently some reviewers (like Gamers Nexus, with the help of software being developed by Intel) are breaking benchmarking down even further to see where exactly during the pipeline the bottleneck is so that you can actually see and chart the impact of driver overhead and CPU speed on GPU render times... really amazing stuff that would have fundamentally changed the development of GPUs, drivers and APIs if it had been available 20 years ago.

... but I digress. FPS numbers are probably fine when we're dealing with frame rates mostly under 40. 😁

Nvidia drivers (in general) don't play nice with some VIA chipsets, particularly the VP3 and MVP3, witch has a negative effect on perfromance. I used a rather early driver, forceware 6.31, witch I've had great results with in the past. This driver seems to be the most tolerant of the MVP3 chipset - however, I believe the low numbers obtained with nvidia cards are due to said driver's high CPU overhead when compared to cards from other manufacturers. I believe I can prove this by benchmarking on a much faster platform, that will eliminate the CPU bottleneck.

As for frame-times, the SiS and S3 were significantly smoother, despite the low FPS. I wish I could see 1% lows in old games, since I believe the stuttering I experienced on nvidia cards is due to a wide gap between 1% lows and maximum FPS, leading to overall "jerkiness". This is most notable with the Geforce 2 MX in Half Life, but it's quite apparent in Homeworld.

Ozzuneoj wrote on 2024-03-18, 07:01:

It's funny we have the luxury of even thinking about such things now. Someone trying to game on a Trident Blade3D was just happy to be able to play a game with hardware acceleration at 25fps back in the day. And if you had bags of money laying around you could get some crazy Quantum3D SLI monstrosity to hit 6ofps at 1024x768 in some games... and two years later it was worthless for any kind of 3D gaming, and your CPU was probably less than half the speed of whatever the latest thing was. 🤣

Things were so much more "interesting" back then...

I remember that feeling - when I got my K6-II in 1999? and fired up quake 2 and I remember it ran pretty decent, even well at 512x460. I completed a lot of 3d accelerated games on that machine. Quake 2, Heretic 2, Blood 2, Homeworld, Kingpin, Dungeon Keeper 1 and 2, Dark Reign 2 and so many others - that little on-board Blade 3D ran like a champ. Looking at the specs for my test card (90MHz core / 90MHz ram) it's easy to see why. Despite having shared memory, the on-board version of the Blade 3D is slightly higer clocked, with memory running at 100MHz, despite being shared, and core running at 110MHz according to documentation. That might explain why my old K6 felt faster - that or the nostalgia of FINALLY being able to play all those demo CD's I'd saved up 😁 I guess if I really want to re-create the experience, I really need to find an MVP4 that runs, or a Trident Blade 3D turbo witch is quite a bit higher clocked (135Mhz(ish) for both core and memory depending on model)

asdf53 wrote on 2024-03-18, 09:54:

Thanks a lot for these interesting benchmarks. I love that you only tested budget cards, to me SS7 is a budget platform so using […]
Show full quote

Thanks a lot for these interesting benchmarks. I love that you only tested budget cards, to me SS7 is a budget platform so using these feels very appropriate. I wouldn't have thought that the Savage4 performs so well and even beats a GF2 MX, especially at these low clock rates. It's also very overclockable, I remember clocking mine to around 150/160 and it didn't even get particularly hot.

The Kyro also performs very well on SS7 systems, I got 3700 points in 3dmark99 with a Kyro 2 (3380 with a GF2 MX). The Geforce also needs old drivers to perform well, some newer games are unplayable with those. The Kyro performs well across the board using the newest drivers, even in OpenGL. I think it was the best performing SS7 card out of all that I tested.

If you feel that your MVP3 board is slow, may I suggest trying these tweaks: https://www.k6plus.com/phpBB3/viewtopic.php?t=1406

On my system, this improved 3DMark99 scores on a Geforce 2MX from 3000 to 3400 points, so it's well worth the effort.

Thank you, and thanks for the link - I'll try some of them on my AX59 Pro. I don't want to mess with this particullar board, as I am quite happy with it's performance.

any visual glitches on sis/s3/trident?

Garrett W wrote on 2024-03-18, 02:10:

No miniGL on GeForce. GLQuake is using OpenGL for which GeForce provides a full ICD. The extensions expose this feature!

afaik no full OpenGL support in GLQuake in the first place, and it doesnt matter if driver offers extension when GLQuake was created in 1997 and doesnt use it.
Looked into the code instead of guessing, GLQuake does indeed use glTranslatef(); glRotatef(); glScalef(); so OpenGL driver can TnL those.

MikeSG wrote on 2024-03-18, 09:16:

It's an AGP 2x motherboard.

WIth AGP 4x, the TNT2-M64 should be +40% faster, and the Geforce 2 MX-400 should be +200-400% faster.

Haha, no! Not even on faster CPUs. You are looking at couple percent at max. Look at platforms memory bandwidth to get a perception of scale, even x2 is too fast for K6.

rasz_pl wrote on 2024-03-18, 12:53:

any visual glitches on sis/s3/trident? […]
Show full quote

any visual glitches on sis/s3/trident?

Garrett W wrote on 2024-03-18, 02:10:

No miniGL on GeForce. GLQuake is using OpenGL for which GeForce provides a full ICD. The extensions expose this feature!

afaik no full OpenGL support in GLQuake in the first place, and it doesnt matter if driver offers extension when GLQuake was created in 1997 and doesnt use it.
Looked into the code instead of guessing, GLQuake does indeed use glTranslatef(); glRotatef(); glScalef(); so OpenGL driver can TnL those.

MikeSG wrote on 2024-03-18, 09:16:

It's an AGP 2x motherboard.

WIth AGP 4x, the TNT2-M64 should be +40% faster, and the Geforce 2 MX-400 should be +200-400% faster.

Haha, no! Not even on faster CPUs. You are looking at couple percent at max. Look at platforms memory bandwidth to get a perception of scale, even x2 is too fast for K6.

Nope. All games tested looked great, with the mention that the Blade3D doesn't do Bump Mapping - but that's a minor detail considering the games it's contemporary with. Output quality on the Trident and Savage 4 is excelent. The SiS 305 on the other hand was over-bright - kind of washed out, like you see on some cheap PCI cards from the mid 90's. Looks fine on a CRT tough. Another mention is that all nvidia cards were very dark under openGL, and no fiddling with the brightness or gamma sliders fixed that. The Savage was pretty dark under openGL as well. The voodoo 2 was too bright in quake 2.