If you want to see my adventure with testing PCI graphics cards and settling on the Radeon 7000,

Your system is beefy enough to bypass driver bottlenecks. Which isn't the case here. Final legacy driver is also horrible idea for pre-Pentium II processor.

Attached is the Savage 4 Windows 2000 driver I downloaded from the S3 legacy driver page back then

The Serpent Rider wrote:

If you want to see my adventure with testing PCI graphics cards and settling on the Radeon 7000,

Your system is beefy enough to bypass driver bottlenecks. Which isn't the case here. Final legacy driver is also horrible idea for pre-Pentium II processor.

I never said it was the ideal hardware or driver for him; it's the one with the most mature OpenGL drivers though.

If he has the Savage4, he should probably go with the Savage4.

appiah4 wrote:

The Serpent Rider wrote:

If you want to see my adventure with testing PCI graphics cards and settling on the Radeon 7000,

Your system is beefy enough to bypass driver bottlenecks. Which isn't the case here. Final legacy driver is also horrible idea for pre-Pentium II processor.

I never said it was the ideal hardware or driver for him; it's the one with the most mature OpenGL drivers though.

If he has the Savage4, he should probably go with the Savage4.

I have some old ATI drivers as well. I'm trying to decide if I have the stamina to test the Radeon7000 and Rage128Pro.

Rawit wrote:

Attached is the Savage 4 Windows 2000 driver I downloaded from the S3 legacy driver page back then

Thanks. I noticed that the folder says "S3_AGP". Was your card PCI or AGP?

Anyone know of a mirror for the old S3 graphics driver website? http://www.s3graphics.com/en/drivers/legacy_s … re_archive.aspx

I'm looking for the Diamond Stealth III S540 Xtreme Drivers for W2K and NT4

https://web.archive.org/web/20021226151637/ht … III_S540_Xtreme
w9x: s540xtreme_win9x_82021.exe , found here: Re: S3 Savage 2000
w2k: s540xtreme_win2k_83024.exe , found here: https://www.driverscape.com/download/diamond- … iii-s540-xtreme
nt4: s540xtreme_nt4_81909.exe , cannot find

Also interested in the non-Xtreme drivers for NT4

https://web.archive.org/web/20021226161612/ht … tealth_III_S540
w9x: s540_win9x_82021.exe , found here: https://www.driverguide.com/driver/detail.php?driverid=75784
w2k: s540_win2k_83024.exe , found here, https://www.driverguide.com/driver/detail.php?driverid=75784
nt4: s540_nt4_81909.exe , cannot find

feipoa wrote:

Was your card PCI or AGP?

PCI. I just downloaded all Savage4 drivers they offered, didn't try this driver myself.

For NT4 I have the attached, didn't try it myself and not sure in what ways it would differ with Diamond drivers.

Got it, thanks. Hard to say if the S3 or Diamond drivers are better. I did find some older Diamond nt4 drivers for the S540 on the disc, but they are not the xtreme ones, which seemed to work best.

Brief update: things aren't looking so good for the Rage 128 Pro. I have a boxed XPERT128, which is a 16 MB version of the Rage128 Pro II. What the heck is a Rage128 Pro II? I installed the drivers from the disc. Ran GLQuake's timedemo, and it hung up after about 2 seconds into it. Rebooted, ran Quake II, screen stays grey at startup. Uninstalled driver 4.13.8002 and installed 4.13.7192, same issue. Installed 4.13.6292, same issue. I couldn't get any older drivers to work.

I pulled the XPERT128 out and replaced it with a Chinese replica Rage128Pro 32 MB, but screen stays blank at power on. Also tried replica Rage 128 VR, but screen stays blank.

I'll try the Radeon 7000 tomorrow. The oldest driver I could find is 4.13.7055.

I really think you'd have a better time with the FX5500 than a doodoo old radeon 7k

Radeon 7000 ran without any issue. I found the oldest driver version which worked with RV100 was 4.13.7199. Only game that didn't play was Unreal in OpenGL mode. D3D worked fine in Unreal. All other D3D and OpenGL games ran. Results were a bit better than the G450 and Savage 4. But now I am curious about which driver versions the frame rates start to drop. Perhaps start with the latest one and work my way older.

I also added Half-Life to the mix of games to benchmark.

maxtherabbit wrote:

I really think you'd have a better time with the FX5500 than a doodoo old radeon 7k

Probably most cards with working T&L would offer some fps for the right game and driver release. Not sure how much more energy I have for this project. There's a dozen games and each card has a dozen driver revisions. Graphic cards up to Q1 2001 (Radeon 7000 or GF2MX400) may be as far as I'd want to take the system. 2004 era cards like the FX5500 or Radeon 9250 start to feel extremely out of place in a Pentium Pro. A PCI GeForce 256 may make a nice fit, but I don't have one of these.

In terms of age of the card per output frame rate, the S3 Savage 4 Pro+ has done fairly well.

if the radeon is performing well with the drivers you're using and you're getting fatigued - I'd suggest you call it good

One cannot know what they don't know; I, therefore, will test at least one or two more drivers with the Radeon 7000. Maybe I've already discovered the most optimal driver, maybe I haven't.

feipoa wrote:

One cannot know what they don't know; I, therefore, will test at least one or two more drivers with the Radeon 7000. Maybe I've already discovered the most optimal driver, maybe I haven't.

Start out by testing the latest, if it isn't slower, you can try something in the middle to see if there are early gains with less overhead.

I run a FX5500 in my pentium PRO but maybe the Voodoo 5500 is a better fit for the P2-overdrive @ 350mhz....

appiah4 wrote:

feipoa wrote:

One cannot know what they don't know; I, therefore, will test at least one or two more drivers with the Radeon 7000. Maybe I've already discovered the most optimal driver, maybe I haven't.

Start out by testing the latest, if it isn't slower, you can try something in the middle to see if there are early gains with less overhead.

That was my plan. The undecided part was which two or three games will be the deciders.

dirkmirk wrote:

I run a FX5500 in my pentium PRO...

Wow, really? For how long now? Which games? I found the GF4MX to be marginally better than the FX5500 on my K6-III-500.

feipoa wrote:

For my system, the solution for the stuttering in GLQuake was to disable dynamic lighting, so r_dynamic "0". For the timedemo, t […]
Show full quote

For my system, the solution for the stuttering in GLQuake was to disable dynamic lighting, so r_dynamic "0". For the timedemo, this only increaes the framerate by 1 fps, or 2 fps if vsync is disabled.

For my system, the solution for stuttering in Quake2 was to disable dynamic lighting, so gl_dynamic "0". For the timedemo, this increased the results by about 15 fps.

For Quake 3, setting r_dynamiclight "0" or using didn't help at all. Using r_vertexLight "1" instead didn't help either. Perhaps the PPRO is just too slow for Quake III?

Is the conclusion that there is a bug with OpenGL drivers for the Matrox G450 w.r.t. dynamic lighting? Anyone know of a fix? From what I could discern, the last version of Matrox's TurboGL wrapper is from driver package 5.52, which does not support the Matrox G450 - only the G400.

Yes you are perfectly right.

I wonder how many Quake benchmarks of the G450 are actually nonsense due to the fact that this bug has only a minor effect on timedemos while the game is actually unplayable. I've put the card into a P3-850 to rule out a CPU/PCI bottleneck and it shows 120fps+ in GLQuake 640x480 timedemo, but the stuttering in areas with dynamic lighting makes it unplayable just as on the Socket 7 board. With dynamic lighting disabled those cards are actually quite decent. Especially since they have DVI and enough RAM to run 1280x1024x32.

I've tried http://dxquake.sourceforge.net/. The DX8 version works for me. While it has generally only half the speed of the ICD, the stuttering bug is still there. So I don't think its actually the ICD, but rather a driver bug. I've build and debugged DXQuake to see how it applies the dynamic light maps. It uses multitexturing to actually apply two textures in one pass (the image and the lightmap). It must relate somehow to the way the multitexturing feature is used by the Quake titles.

I've already tried TurboGL with the modded driver CD here at Vogons. It installs fine, but it doesn't look as if it makes any difference. It is running though, because Quake2 will lock up once installed. I'll try modding some old driver .inf to support that card and see if those make any difference. The last hope may be to install older Bios versions. It would be nice to know Bios dates and versions of someone with G450 AGP cards that don't show the bug.

Edit: Just tried GLQuake with command line option "-nomtex". The stuttering is gone, even with gl_flashblend=0, and without any apparent quality loss. FPS drops from ~120 to ~80, though.

feipoa wrote:

dirkmirk wrote:

I run a FX5500 in my pentium PRO...

Wow, really? For how long now? Which games? I found the GF4MX to be marginally better than the FX5500 on my K6-III-500.

why is that a surprise 🤣, I've been saying I run a FX5200 in my PPro for the whole thread now

Surprised because of the age disparity. I usually see this done just for testing purposes. There is, for one reason or another, disapproval for mixing hardware of significantly different generations.

watz wrote:

Edit: Just tried GLQuake with command line option "-nomtex". The stuttering is gone, even with gl_flashblend=0, and without any apparent quality loss. FPS drops from ~120 to ~80, though.

That is a dramatic drop in output. How did Quake 2/3 make out? Is there a parallel command in these games?

watz wrote:

Edit: Just tried GLQuake with command line option "-nomtex". The stuttering is gone, even with gl_flashblend=0, and without any apparent quality loss. FPS drops from ~120 to ~80, though.

I could live with that. Do you know if there is a similar switch for Quake 2 or 3 (not that I had issues with Quake 3 to be honest)

I'll test out this -nomtex option once I'm finished with the Radeon 7000 drivers.

With -nomtex added to the shortcut startup command, the framerate drop w/Vsync enabled is from 43.6 fps to 40.6 fps. Or from 55.8 fps to 50.9 fps w/Vsync disabled. The -nomtex flag does get rid of the mild stuttering when using the G450.

I also tested the G200 using the same driver version of 6.83 and want to point out that the G200 does not suffer from the same stuttering bug as the G450.

Concerning the Radeon 7000, the oldest driver I tested is the fastest, which is 4.13.7199. The newest driver, 4.15.9165 works, but frame rates drop by a small amount in most games. For example, GLQuake drops from 71.1 fps to 59.9; Quake III drops from 19.6 fps to 17.5 fps; Unreal drops from 20.7 fps to 17.4 fps; Turok2 from 29.3 fps to 23.1 fps; the Quake 2 score did not drop.

By the way, in case anyone was wondering how the Oxygen VX1 compares with other PCI cards on a K6-III 500, I ran some benchmarks with charts here:
Re: Best PCI VGA for a K6-III+ Cards compared are the GF4 MX440, FX 5500, GF2 MX400, Voodoo Banshee, Matrox G450, Oxygen VX1, Rage 128 VR, TNT2 M64, and TNT.