The TurboGL driver is without a doubt a bit faster.... on my K6-2 380 with G400 using Powerdesk 5.52 drivers and TuboGL the differences are

Without TurboGL
Quake2 - 28FPS
GLQuake - 48FPS

With TurboGL
Quake2 - 35FPS
GLQuake - 57FPS

Needs to be noted that they're not entirely stable though.... the TurboGL drivers even caused a system hang in Quake2

Well, that's basically 20-25% faster, pretty big difference really..

Next it needs to be shown that relative to the new drivers the older drivers don't simply trade image quality for speed, i.e. that if there's a difference in speed in favor of the older drivers, you couldn't get the same performance boost at the same level of visual fidelity on the newer drivers by lowering the in-game/driver detail settings.

Dirk Daring wrote:

The TurboGL driver is without a doubt a bit faster.... on my K6-2 380 with G400 using Powerdesk 5.52 drivers and TuboGL the diff […]
Show full quote

The TurboGL driver is without a doubt a bit faster.... on my K6-2 380 with G400 using Powerdesk 5.52 drivers and TuboGL the differences are

Without TurboGL
Quake2 - 28FPS
GLQuake - 48FPS

With TurboGL
Quake2 - 35FPS
GLQuake - 57FPS

Needs to be noted that they're not entirely stable though.... the TurboGL drivers even caused a system hang in Quake2

VIA chipset there? Powerdesk 6 is said to be more stable with VIA. Powerdesk 6, however, doesn't work with TurboGL.

Is your G400 running AGP 1x? I believe with VIA you want it at 1x, though I don't know if that will make it completely stable. There's an official Matrox Technical Support Utility that has compatibility options like AGP speed.

swaaye wrote:

Dirk Daring wrote:

The TurboGL driver is without a doubt a bit faster.... on my K6-2 380 with G400 using Powerdesk 5.52 drivers and TuboGL the diff […]
Show full quote

The TurboGL driver is without a doubt a bit faster.... on my K6-2 380 with G400 using Powerdesk 5.52 drivers and TuboGL the differences are

Without TurboGL
Quake2 - 28FPS
GLQuake - 48FPS

With TurboGL
Quake2 - 35FPS
GLQuake - 57FPS

Needs to be noted that they're not entirely stable though.... the TurboGL drivers even caused a system hang in Quake2

VIA chipset there? Powerdesk 6 is said to be more stable with VIA. Powerdesk 6, however, doesn't work with TurboGL.

Is your G400 running AGP 1x? I believe with VIA you want it at 1x, though I don't know if that will make it completely stable. There's an official Matrox Technical Support Utility that has compatibility options like AGP speed.

SIS 5591, and AGP 1X is all it will manage, both the board and card claim to be capable of 2x, but there seems to be no way to enable it.
With Powerdesk 6 there are instabilities with GLQuake..... FPS will just drop down to nothing in areas with certain types of lighting

Dirk Daring wrote:

SIS 5591, and AGP 1X is all it will manage, both the board and card claim to be capable of 2x, but there seems to be no way to enable it.
With Powerdesk 6 there are instabilities with GLQuake..... FPS will just drop down to nothing in areas with certain types of lighting

Interesting! The joys of Super 7. The only board I still have is the ASUS P5A with Aladdin V. It has AGP problems of course but never tried a Matrox AGP card on it. I usually only consider 3dfx cards on any AGP 2x era non-Intel motherboards.

Interesting. If I can find the energy, I'd like to do some tests with the TurboGL drivers soon. I got my G400 MAX in a K6-3 550MHz at the moment, and I haven't really benchmarked it recently. I remember doing some tests a few years ago, comparing how it performed in different setups, going from K6-2 to PIII 1400, and seeing an amazing scaling with CPU power. In fact, at low res (640x480) it ended up being able to beat it's more powerful sibling, the Parhelia. Sure, the Parhelia would probably perform better in a P4 or Athlon XP, but I don't currently have the hardware to check that.
I know my bf has a P4 motherboard and CPU sitting idle in his basement though, so I might as him to bring it over so I could do some testing. We shall see. I suffer from depression and anxiety, so I find it really difficult to find the energy to actually do stuff, even when it's things I enjoy.

Different drivers can give MASSIVE gains, that's why it is so hard to compare scores between different machines with different hardware and different software...

Good example is the ''aiming for the stars'' thread here on Vogons where we try to reach record scores with k6-3+ cpu's in 3dmark2000. link AMD K6 3DMARK, aiming for stars.

There you can see scores from a k6-3+ with a geforce2 or geforce3 in 3dmark 2000 of around 5000 3dmarks (not talking about cpu marks). Greatly because of the 7.76 driver of Nvidia.

To be able to beat that score with a much faster card (like a geforce4 ti or even 5950ultra, 6800ultra or faster) i need a fast Tualatin cpu or higher just because of the later driver beeing much slower in 3dmark 2000 and the Nvidia 7.76 driver is not supporting those newer cards.

I recently did some benchmarking on my G400Maxx to compare it to my Matrox Parhelia and figured I would post it here. I benched all the G400 drivers I could find for windows xp. I'll do the same thing with windows 98 in my 1GHz p3 system too, once I'm done with the Parhelia. Here is the link: https://docs.google.com/spreadsheets/d/1OydTR … dit?usp=sharing. I did the benchmarks on my 4coredual-vsta board with 4gigs of ddr2@533 and a core2 e6700. r_picmip=1 is the default in quake3 and that is probably what Phil and Anand benched with. I also included benchmarks with max settings and the Parhelia cards

Weren't drivers 'optimized' for the Futuremark tests though? You really need to be using a less well known benchmark to test these things.

The dropoff after 584 is rather interesting, that is when they dropped TurboGL and moved onto a full OpenGL driver I suppose?

I think that over time Matrox just focused purely on driver stability as opposed to performance, as they sold (and still do sell!) these cards for basic display systems, such as signage and basic dual display output. Under Windows XP, I found that the whilst the 5.93.009 drivers are the last to offer Direct3D support, they cannot render into textures, causing 3Dmark 2001 to refuse to run! The previous 5.92.006 drivers do work, but I would be very interested to see how much better even earlier drivers perform.

My guess is TurboGL is the way to go for Quake based games. 😀

I wonder if it's possible to use TurboGL with G450.

Ok. I did some testing on Windows 98 with a P3@1GHz, 256MB at PC133 CL2 and a G400Max. The TurboGL is only supported on Win9x so my earlier tests were all with the full ICD. Matrox recommends an AGP aperture size of 256MB but there is no option for it in my boards BIOS and powerstrip says its set at 64MB. Matrox says there may be texture corruption with a smaller aperture setting and there is in UT99. Although quake3 looked fine. Its an intel 815EEA board, guess I need to look for an update. Anyways, results are here: https://docs.google.com/spreadsheets/d/1TB1Cv … dit?usp=sharing. I can put in my G450 and see if I can get the TurboGL to work with it.

Edit: Forgot to mention. The 552 drivers gave quake3 a 16bit z-buffer and an 8bit stencil. All the other drivers just gave quake3 16+0. My Radeon 8500, Geforce2 and Parhelia give it 24+8. My Geforce4 and FX give it 24+0.

After a bit of trial and error I got the 1.3 version of the TurboGL to work on the G450. I had to install the driver off the G450 driver CD that I got from the Vogons driver library. It's version 603, date is 2000/08/17. The performance improvement was small though, went from 63 fps at 640x480x32 to 63.3. There is probably a bigger improvement with 16bit colour or maybe a G450 paired with a much slower cpu like a K6-2. Id have to do more testing to find out.

Based on your testing the 2000/XP drivers for the G400 seem to have a nasty bug starting with 5.84 that halves 32bit performance, something that does not exist in your P3/Win9x testing (where performance actually improves..) I would like to know what is going on there. Although Phils testing was on Win98 he may have also encountered that bug and maybe it is not OS specific but is related to BIOS settings like aperture size or maybe left over files from earlier drivers or what not? Please investigate!

RichB93 wrote:

I think that over time Matrox just focused purely on driver stability as opposed to performance, as they sold (and still do sell!) these cards for basic display systems, such as signage and basic dual display output. Under Windows XP, I found that the whilst the 5.93.009 drivers are the last to offer Direct3D support, they cannot render into textures, causing 3Dmark 2001 to refuse to run! The previous 5.92.006 drivers do work, but I would be very interested to see how much better even earlier drivers perform.

Interesting. I didn't realise Matrox cut out their Direct3D support in drivers. Is this only on W2K/XP drivers, or did the Win9x get Direct3D cut as well? If so, which driver versions are the last with Direct3D? Was OpenGL cut out as well?

I remember when I did some tests with my K6-III-500 system using PCI cards, the G450 couldn't run Final Reality (which I believe is DirectX5) and I was using driver 6.82 (latest). The G450 also scored poorly in Quake II - 20 fps at 800x600. Even the Rage 128 VR scored 40 fps on the same board. Oddly, in GLQuake, the G450 did a lot better with 45 fps, compared to 51 fps for a GF2 MX400. I suppose I should have played around with the driver versions like you are doing. I normally test all driver versions with GeForce cards to find the fastest, but for some reason I didn't do this with the G450. It would be nice to have a list of the generally most optimal driver versions for the G-series cards for Quake1/2/3.

cxm717 wrote:

Ok. I did some testing on Windows 98 with a P3@1GHz, 256MB at PC133 CL2 and a G400Max. The TurboGL is only supported on Win9x so my earlier tests were all with the full ICD. Matrox recommends an AGP aperture size of 256MB but there is no option for it in my boards BIOS and powerstrip says its set at 64MB. Matrox says there may be texture corruption with a smaller aperture setting and there is in UT99. Although quake3 looked fine. Its an intel 815EEA board, guess I need to look for an update. Anyways, results are here: https://docs.google.com/spreadsheets/d/1TB1Cv … dit?usp=sharing. I can put in my G450 and see if I can get the TurboGL to work with it.

Edit: Forgot to mention. The 552 drivers gave quake3 a 16bit z-buffer and an 8bit stencil. All the other drivers just gave quake3 16+0. My Radeon 8500, Geforce2 and Parhelia give it 24+8. My Geforce4 and FX give it 24+0.

On your spreadsheet what are the two different Parhelias? 256? I thought the Parhelia was 512bit, what do you mean?

mzry wrote:

cxm717 wrote:

Ok. I did some testing on Windows 98 with a P3@1GHz, 256MB at PC133 CL2 and a G400Max. The TurboGL is only supported on Win9x so my earlier tests were all with the full ICD. Matrox recommends an AGP aperture size of 256MB but there is no option for it in my boards BIOS and powerstrip says its set at 64MB. Matrox says there may be texture corruption with a smaller aperture setting and there is in UT99. Although quake3 looked fine. Its an intel 815EEA board, guess I need to look for an update. Anyways, results are here: https://docs.google.com/spreadsheets/d/1TB1Cv … dit?usp=sharing. I can put in my G450 and see if I can get the TurboGL to work with it.

Edit: Forgot to mention. The 552 drivers gave quake3 a 16bit z-buffer and an 8bit stencil. All the other drivers just gave quake3 16+0. My Radeon 8500, Geforce2 and Parhelia give it 24+8. My Geforce4 and FX give it 24+0.

On your spreadsheet what are the two different Parhelias? 256? I thought the Parhelia was 512bit, what do you mean?

It's a 256MB Parhelia. Its clocked at 250MHz. The 128MB Parhelia is clocked at 207MHz. I ran some benchmarks on the 2 Parhelias and the 256MB one is much faster, sometimes as much as 40%. I'm not sure why as its only clocked 20% higher. Here is a link to those tests I ran if you're interested https://docs.google.com/spreadsheets/d/1l0S2w … dit?usp=sharing