swaaye wrote:

G400 Max is faster than Voodoo3 3500 and TNT2 Ultra with enough CPU behind it (say a P3 600). It's entertaining to see Voodoo2 beating it. Good work, AMD K6-2.

I think that's the norm for Matrox cards. They need a lot of CPU power to come to their full potential.

LunarG wrote:

I think that's the norm for Matrox cards. They need a lot of CPU power to come to their full potential.

They are a little less efficient yes, but if you compare cards on a P3 500 or so the GPU chart will look right. With the K6-2 the results are entirely CPU bound.

LunarG, do you change the settings manually each time you need to rebenchmark a game? After a while it gets quite tedious, i think i'm going to create a batch for each setting in the chart that basically deletes either unrealtournament.ini or q3config (depending on the game you're benchmarking), copies a already made setting and runs the game afterwards. It will save a lot of time!
I mean for me it does since i need to run the same benchmark for three different drivers each with vsync on and off.

leileilol wrote:

LunarG wrote:

I got Q3 sorted today, had to find the right patch to make it run on Windows 98se. For some reason it won't let me swap from "default" rendering and to Voodoo. Not sure why.

Try manually setting r_glDriver to 3dfxvgl (or 3dfxogl in some cards).

All versions of Q3 should work on 98se. Ioq3 doesn't and would need KernelEx to run (because of PSAPI usage and Winsock2 getprocaddr usage making it require XP at the minimum) - and by the way, that engine removes r_glDriver, WGL_3dfx_gamma_control and vendor-specific fixes so it's not really optimal or recommended on older hardware - so I hope you're not using Ioq3.

I can't remember the error message now, but it was something about "found interface version 4, expected 3" or some such. A 1.32 something something patch sorted it though.

Yeah, that has to do with the protocol change introduced in 1.25-1.27 in Sept-Dec 2000, and running a pre-1.25 exe with 1.25+ PK3s.

on a p3s-1.4g, the g400max can match or edge out tnt2ultra on some well-optimized games like quake engines. but when it comes to 3dmark its still no match. my guess is that matrox could only optimize their drivers for certain games.

noshutdown wrote:

on a p3s-1.4g, the g400max can match or edge out tnt2ultra on some well-optimized games like quake engines. but when it comes to 3dmark its still no match. my guess is that matrox could only optimize their drivers for certain games.

If my Compaq OEM "TNT2 Pro" is genuinely a TNT2 Pro, it shouldn't be that much slower than a TNT2 Ultra, at least not according to the old reviews I found online recently. If that's truly is the case though, I can only say that on a sufficiently fast CPU (P3EB 800MHz) the G400MAX totally makes mincemeat out of the TNT2 Pro. Back when these cards were still fighting it out for the throne though, I remember the TNT2 Ultra usually winning the OpenGL benchmarks and the G400MAX winning the Direct3D benchmarks, although the results were very close. Very late in the life of these two cards, when Matrox finally got their shit together, the G400MAX also got good OpenGL performance, but by then nVidia was just about to release the GeForce256 anyway, so it didn't really make much of a difference.
Perhaps I'll have to keep my eyes open for a TNT2 Ultra, just to be able to do some comparison benchmarks at some point 😉

Firtasik wrote:

leileilol wrote:

if you use r_primitives 2.

I've discovered recently that Wolfenstein: Enemy Territory is running slow on my Radeon HD 5770 without this command.

Matrox G400 supposedly has a similar issue with compiled vertex arrays not being used by default requiring both r_primitives 2 and r_ext_compiled_vertex_array 1...

Later at night i'm going to post another chart comparing results between 3 different nvidia drivers. This time around i've got much less performance on UT in high detail mode, i really don't get why. Weird stuff.

Rick6: I would like to thank you for sharing your benchmark results here! I was doing similar tests (on much smaller scale) on my K6-2/550MHz, I needed some results for comparison and here they are! 😀 Hopefully I will add Matrox G400 in a few days...

sledge wrote:

Rick6: I would like to thank you for sharing your benchmark results here! I was doing similar tests (on much smaller scale) on my K6-2/550MHz, I needed some results for comparison and here they are! 😀 Hopefully I will add Matrox G400 in a few days...

You're welcome!
And heres a comparison between the driver 28.32; 30.82; 44.03 on the geforce256 sdr in a amd k6-2 at 500mhz:

To be honest, so much work and really not much to see here because performance seems to be about the same across the board. Apart from the 3dmarks benchmarks of course.
Odd thing here is why in the world i'm having worse performance on ut in high quality than in my first attempt. Beats me, maybe i should reinstall everything.
I was only able to disable vsync at driver level on the driver 44.03 (opengl). For UT i disabled it in it's *.ini file (Direct3D).

rick6 wrote:

I was only able to disable vsync at driver level on the driver 44.03 (opengl). For UT i disabled it in it's *.ini file (Direct3D).

to disable vsync on old drivers(30 and older), you need to add the infamous "coolbits" registry, which also enables overclocking options. you will then see vsync option in "additional settings" in d3d page, and vsync option in opengl page is always available.

there should be no difference between 28.32 and 30.82 as they belong to the same 20 series, this series just had so many minor updates that the driver number went above 30.

between 30 and 40, there was a significant >5% boost in 3dmark if you have a very fast cpu. but as i said, newer drivers consumes more cpu so it would be even slower if your cpu is slow, as you can see in 3dmark01.

noshutdown wrote:

between 30 and 40, there was a significant >5% boost in 3dmark if you have a very fast cpu. but as i said, newer drivers consumes more cpu so it would be even slower if your cpu is slow, as you can see in 3dmark01.

The conclusion would be that newer drivers will improve direct 3D performance on games\benchmarks that are not much cpu dependant. That should be why i did not notice any improvement on UT and the high detail car chase benchmark in 3dmark 2001 which are way too much cpu dependant!

I was planning on running some more benchmarks on my system, but it's been put on hold due to the fact that I've just had a tumor removed from my leg. Means I've been ordered to sit with my leg up for the next couple of weeks. Hopefully I'll be able to get some benchmarks of my Radeon 7000 soon.

Sorry to hear that 😒 Rest up and get well!

Hope you have a speedy recovery friend!

Gah! The HD I was using for the K6-2 system seems to be faulty. Well, I knew something was wrong with the fact that it's labelled 20GB but only gets detected as 10GB (it did in the Compaq I took it from as well), but today it stopped wanting to boot my system. If I try to auto-detect a Seagate Barracuda 7200.1 80GB drive from the BIOS the system just hangs, so I've put on the capacity limitation jumper to set it to 32GB. This seems to be the solution for this system. I'm installing Windows 98se on it again at this very moment. Hopefully this system will be running better from now on.

Right. System is running fine, so I tried my Radeon 7000 64MB DDR last night. System booted up, I installed drivers and rebooted. Dxdiag reports that there's no direct draw or direct 3d device in the computer. 3DMark99MAX says I need a DirectX6 compatible device. Dxdiag also claims I only have 5MB of graphics memory (or "unknown" if I try a different driver build). The ATI control panel does report 64MB though. I will try the card in my Pentium III to see if I get the same results there. I worry that my AOpen AX59 Pro just doesn't like the Radeon card.

Any update yet? 😀

LunarG wrote:

Back when these cards were still fighting it out for the throne though, I remember the TNT2 Ultra usually winning the OpenGL benchmarks and the G400MAX winning the Direct3D benchmarks, although the results were very close. Very late in the life of these two cards, when Matrox finally got their shit together, the G400MAX also got good OpenGL performance, but by then nVidia was just about to release the GeForce256 anyway, so it didn't really make much of a difference.
Perhaps I'll have to keep my eyes open for a TNT2 Ultra, just to be able to do some comparison benchmarks at some point 😉

i would say the situation today is simply opposite: g400max can match the tnt2ultra in opengl tests, but in 3dmark it got smoked. on my pt880 rig the tnt2ultra managed to hit 2350pts, while g400max struggled to reach the 2000 barrier.
just think: as quake3 was released in 1999, the 3d support required for it are actually pretty simple compared to the 3dmarks. thats why matrox could manage to optimize their opengl drivers, but not the support for newer d3d benchmarks and games.