View Full Version : VERY (I mean very) disappointing results with Q3 and TurboGL

10th October 1999, 11:23
Fist of all, my system:

P3 450 oc 600 (133x4.5)
QDI BrillianX 1S
128mb SDRAM
G400 32mb DH oc 165/206 (with MgaTweak)
SB Live basic
Windows 98 with SE patch (Italian version)
DX 7 (final English version)
Driver ver 5.30
Monitor: 1024/85hz--800/85hz--640/100hz

I obtain very disappointing results with Q3Test 1.08. I reinstalled the demo, and the drivers several times (with pd uninstall). I tried to change video bios/shadow settings in bios. I tried not to overclock the card. I tried also to change Powerdesk registry settings with G400Tweak. I tried to change deskop colour depth/resolution. I tried all of above with a fresh Win98 installation (I have it on my 2nd HD; I use it only for watching DVDs).
That's my results (timedemo 1 - demo q3demo1.dm3). They are obtainded at 32bit color, with 16bit z-buffer, and everything to max (even textures quality an trilinear filtering). Sound is low. Game option all on, except for v-sync (I controlled in "q3config.cfg" too) and flare.

1024 28,1 28,4
800 41,6 43,1
640 47,3 62,5

I also tried to run test with vsyc on and TurboGL. I obtainde the SAME results as v-syc off (that's very strange). For that reason, I tried to change 800x600 refresh from 85hz to 100hz (I thought maybe v-sync can't be disabled): same results. At last, i tried to install the previous beta driver (the one that come in G400icd.exe). I obtained:

1024 28,7
800 43,6
640 51,7

Compare with the previous ones...

My TurboGL version is (is the same than yours?)
My g400icd.dll vesion is 5.30.006

Why are it so LOW? Why no differences with v-sync on/off? I can't get it.
I tried also Q2, and I obtainde 110fps in 640x480x32. I think that result is correct.
Can you help me? Have you an explanation?

tish beta2
10th October 1999, 16:00
My experience is very much the same as yours. Better, but not as much as the reviewers got. This looks like the 5.21 "promised" OGL improvement. The performance increase was incredible, but only with a certain Beta build.

I don't think they'll release that beta as they did last time...

10th October 1999, 18:34
Same low scores here . At 1024 32bit (all on max settings) i cannot break the 30 fps barrier with my g400max and PIII540.
I donīt think that there is a special beta TGL built that ant was using rather than it was not able to enable all features like trilinear etc.
Of course i really hope that there is a special TGL beta that really boosts performance to mid 40īs fps in 32bit.

ANT , please bench with the public TGL driver !!!!!!!

tish beta2
10th October 1999, 18:41
He will. I thought he also mentioned that he might have had trilinear filtering disabled, that would probably explain it...

10th October 1999, 19:14
Same here : turbogl way underperforms in q3demo1.

First off , the Turbogl driver makes 1.5 fps difference on this machine, P600 @ 10 x 7 with everything high/32bpp/trilinear/sound high but textures only at 3/4 . Luckily that's an increase. From 31.5 to 32.9 fps. I'm underwhelmed.
This result does not vary at all with no sound!
Unfortunately , with textures at 4/4 ,the demo is unplayable, ~12 fps.

Secondly , I have to disable 32 bit Z buffer or I lose 3 fps .

'Sup ?

In Harm's Way

[This message has been edited by troop (edited 10-11-1999).]

11th October 1999, 01:18
im surely not sure but i think some of the reviewers if not all tested with audio disabled.

tish beta2
11th October 1999, 09:15
I tested with and without sound. The difference was 0.4 fps.

12th October 1999, 01:04
One thing to remember:

The only really significant increases I've seen anywhere with TurboGL have been at low resolutions. Even with the old drivers, the G400s were very competitive at high resolutions in OpenGL. What surprised everyone was that the G400s could kick everybody's butt at 1280x1024 and 1600x1200, but would fall back by 30-40 fps at 640x480. The old drivers were pretty good at high resolutions and horrible at low resolutions. Well, the TurboGL drivers have fixed the problem at low resolutions. They don't seem to have much of an effect at high resolutions because the drivers were already pretty good at high resolutions.

The biggest complaints I've seen have been people running at 1024x768 and wondering where their 30-40% speed increase went. Well, the 30-40% speed increase was back at 640x480. This is backed up even in this thread, where Sharkyz got a 32% increase with the TurboGL drivers at 640x480.

I guess my point is that the TurboGL drivers aren't the final version of Matrox's OpenGL drivers (Thank God!) and that speed will hopefully rise with each driver release.

Now, with that said, I get to rant. What about NT? Those of us who actually use these cards for business purposes (and/or want dual processors for other reasons, ahem, games) are suffering with horrible 32-bit performance in OpenGL(Dual P3-500, 256MB PC100 RAM, Millenium G200 8MB, Quake 3 640x480x32 - 6.4 fps) and image quality issues to boot. Maybe I just need have a little patience myself and wait for Win2000 (God help us!).

If you read all this, I apologize for the length and my ranting,


[This message has been edited by noackjr (edited 10-12-1999).]

12th October 1999, 02:21
Hey Troop, I remember talking to you before about Q3test scores. I am now get about 25 fps with max everything, including texures. Check your AGP using PCIList to see if everything is working right.

12th October 1999, 05:02
Good point Gurm. I've noticed the same thing here on my P3-450 (at 558). Q3Test1.08 is playable with everything on (including trilinear) at 1024 now- everywhere in all the maps. This includes the 2 areas that used to slow me to slideshow rates with heavy fire before in map 1 - in the portal room, and in the area where the Quad damage appears (when facing toward the exit to the railgun area). However, with trilinear on, my framerate appears to average 27 FPS on this map, only 3 FPS more than with the beta GL. With bilinear, I get as much as 37 FPS in 1024 on this map.

Major difference!

It strikes me as kind of strange that I never noticed much of a difference between trilinear and bilinear filtering, either in visual quality, or in performance, with previous drivers. With the latest drivers, I see a major difference in both (the images are much more vibrant seeming, for one).

All in all, this seems to be a major step in the right direction for Matrox.


[This message has been edited by Ace (edited 12 October 1999).]

12th October 1999, 06:33
Ozy---The AGP is x2 by PCIlist and I checked 3DMark99MAX --the 32MB texture speed is 82 fps, better than no AGP. The test score here is ~5700 marks with 32 bit Z on. About average I think for the cpu.

I don't follow the idea of disabling features in this situation , eg trilinear filtering--we are following published test settings --I understand that all features for the Q3A published test results , including Ants' , are 32bit/trilinear/high including textures.Sound or not does not matter.

Why change those ? The logic of that is "turn off all features and get blinding speed !"

The troubling test result here is the textures. With them turned all the way to the right , demo q3demo1 is in a trance--~12 fps. With it at 3/4 , I score a very playable 32 fps , @ 10v7x32 on the desktop and in the demo. This is a 110Hz refresh . This score is similar to Sharkyz above but he has full textures and runs at 28 fps .

32bit Z B costs 3 fps here but is not really a test setting as no reviewer appears to have specified it ---its a video card setting.

From what I see , some machines can run the demo at full features--eg , lightmapping , and full textures, 10x7x32bit on an 19 " display and score over 35 fps. I'd really like to know how .I wanna buy some of that .

In Harm's Way

[This message has been edited by troop (edited 12 October 1999).]

12th October 1999, 07:13
Like I said, with everything all the way up I score between 23-25fps, depending on the Zbuffer. Hq sound, trilinear, full textures, 32bit color and textures, etc. Now I see reviewers getting over 30fps with these same settings at 1024x768 and I wonder how, just like you. However, with the TurboGl in terms of playability as opposed to benchmarks I have seen a huge difference. Over 100fps in 1024x768 at times. The G400 is currently the performance leader in OpenGl, over both the GeForce and the TNT2Ultra, at least at 1280res and 32bit color depth.

12th October 1999, 09:42

If you have a P3, and you're still getting scenes where it slows to a crawl with those settings, I'd be suspicious that TurboGL isn't installed. The new full ICD turns out numbers like you're reporting. Please verify for us that you're actually using the TurboGL...

As for disabling features - Ant's tests ran with an earlier version of the TurboGL. This SEEMS to be the first version that actually implements Trilinear filtering. Therefore, to get numbers like the others get you have to disable Trilinear. That's all.

- Gurm

G. U. R. M. It's not hard to spell, is it? Then don't screw it up!
The word "Gurm" is in no way Copyright 1999 Jorden van der Elst.

12th October 1999, 10:10
Absolutely I use the Turbo. Using the full ICD reduces the fps by only a marginal amount anyway--1.5 fps.I use a PIII 600, 128 MB , etc.
If Ant and the others can confirm that the results were achieved using bilinear filtering I will rerun the test. It will be faster I'm sure to run the textures at the reduced level of filtering. But is that what the published results reflect ? I'd like that confirmed before I change the test conditions. Otherwise , or even still ,it appears that I have a problem *or* there's some good stuff around.

In Harm's Way

12th October 1999, 16:42

Read elsewhere on these forums. Manoj figured out where the speed decreases were coming from - Trilinear (30% speed hit) and 32-bit Z-Buffer.

Try it without those. Trilinear is pretty ugly in Q3, _ALL_ cards take a 30% hit except the TNT which doesn't really do Triliniear anyway, it dithers.

- Gurm

G. U. R. M. It's not hard to spell, is it? Then don't screw it up!
The word "Gurm" is in no way Copyright 1999 Jorden van der Elst.

12th October 1999, 16:49
The only thing I can figure which explains the difference in scores on similar cpus is screensize. I can add 20% to a test result by running at less than full screen on this 19"
Hitachi 751. Maybe Ant and Anand are running high cpus on puny screens ?

In Harm's Way

tish beta2
12th October 1999, 19:16
I just wanto whine a little here with Jon on the winNT subject. How long has there been an OGL ICD out for the G200, and how much has it improved? If Nvidia can do it, why not Matrox? Aren't they as good as people say? Now even reviewers put up the good old "OGL ICD for win98, winNT and win2K" sentence when they talk about any TNT cards (1, 2, geforce etc).

I would have fired this up really good in a desperate effort to catch Matrox' attention and get them to work harder on that thing, hadn't it been for a careful warning from Ant http://forums.murc.ws/ubb/wink.gif. Hey, we wanted an option utility for the G400, we got it. We wanted the good Beta ICD and we got it. Maybe if we whine and scream loud enough in these forums.... http://forums.murc.ws/ubb/wink.gif http://forums.murc.ws/ubb/wink.gif

Anyway, it's good the TGL thing is sorted out (thanks to Manoj). I guess the beta TGL Ant benched with had those things disabled.

5.21: reviewers get a beta ICD that gives 30% increase. We get the "final" ICD, and get dissapointed.

5.30: reviewers get a beta TGL that gives 40% increase. We get the "final" TGL, and get dissapointed.

See the pattern? big M is a sly devil http://forums.murc.ws/ubb/icons/icon6.gif here... most folks will give their attention to the reviewed beta drivers, and only discover the truth if they actually has the card for themselves.

Thanks + Best Regards

12th October 1999, 22:42
Are you stupid tish? Why would the reviewers get a TurboGl that was so much better? Damn that Matrox...(insert paranoid rant)...they're always...(insert useless comment)...enough said!! I get big improvements in gameplay from the TurboGl, so I am quite pleased with it.

13th October 1999, 08:48
What would be nice is for MURC to redo the tests under simple conditions with fixed parameters. I suggest Q3demo1 @ the " High "
quality level as set by the demo and the only variation from all default settings is in resolution.For the video card itself , 32b ZB is off.

Along the lines of the suggestion that somehow the original test was run on a bilinear filter with 32b ZB off , I notice that on this system , those are the default conditions . After the driver is updated , 32 bit ZB is not checked here in display properties,though I suppose it could be enabled if the update wre to copy current settings from an installed powerdesk. But most users now uninstall first and it appears that the default condition is 32b ZB off.

Next I noticed here that the Turbogl defaults to bil filtering in Q3 on installation. I have to reset tri filtering to get that when I reinstall the Turbo.
SO ,did the published testers not examine the settings in the demo and for the card to specify these items after installing the 5.30 and Turbo ?
Can MURC redo the PIII 600 Turbo test ?

In Harm's Way

13th October 1999, 16:22
As far as this thread is concerned , nothing
has been figured out . There is only a suggestion that the Z buffer and Trilinear filtering may have been set to off for the TGL results published everywhere which seem unattainable by consumers. We are waiting for further information on these points. The screen size is also important as one can see
when that size is varied between tests.

In Harm's Way

tish beta2
13th October 1999, 17:33
<self-defense mode>

Why would the reviewers get a TurboGl that was so much better?

Well, they did, didn't they? Like I said, if Matrox give out an early beta of some drivers that disable features to get things faster to reviewers, folks that read their reviews will say "wow that OGL has really improved!"

I too get improvements from the TGL, just like I got improvements from the 5.21 driverset. But not what the reviewers got. I read two reviews of the TGL driver that said it gave 40% performance boost in 1024@high in Q3, but I got 5%. And I had the excact same system as the reivewer!

Ozymandis, I'm not saying Matrox is always useless low-lifes (like it appeared as I felt in your post), I'm merely exeviating my thoughts around the OpenGL performance in windows98.

You may be pleased with win9x OGL performance. Ok, so am I. But I'm not pleased with what the dissapointments I've had, and the winNT ogl situation.

</self-defense mode>

As far as this thread is concerned , nothing has been figured out

According to one of the BetaBoyz (tm), it has already been;

Quoting Gurm

Read elsewhere on these forums. Manoj figured out where the speed decreases were coming from - Trilinear (30% speed hit) and 32-bit Z-Buffer.

Thank you + Yours Sincerely

Sean Segel
13th October 1999, 18:22
Add to that list:

Turning the Texture Quality slider to the max contributes to lower framerates.



tish beta2
13th October 1999, 22:02
No problem.

Hey, now I get exact same results as Ant did! +/- 1 fps! Hmmmm...... then pherhaps Texture Quality, 32 bit Z-Buffer AND trilinear filtering were disabled in the drivers Ant got to preview!



I guess we sorted that one out too...

14th October 1999, 05:10
Absolutely not---the MURC review of the beta Turbo was @ 6oo MHz with "trilinear filter" and no mention of ZB.
Feeling sleepy?

In Harm's Way

14th October 1999, 07:03
The Murc review was @ 500 and with sound disabled.


14th October 1999, 12:25
Yup--sorry 'bout that. @500MHz --and he scored over 40 fps at High quality with tri filter and max textures ??--I get 33 fps on a P600, 19" screen, 110Hz, with 3/4 textures.

In Harm's Way

tish beta2
14th October 1999, 20:36
Troop you misunderstood me

Everybody know that Ant's preview of the TurboGL was tested on quake3 with High Quality Settings. It's obvious when we look at the other scores with the Beta ICD.

What I said was, there is a conspiracy theory that trilinear filtering and 32-bit z-buffer was actually completely disabled in the TGL drivers for Q3. Since that driver is a miniGL developed spesifically for four games, It's not hard to just cut away that special feature for that game...

I have a P3 450@504 and 192 mb ram, the Max (not overclocked) and clean win98se install, and my normal setting was exactly Ant's TGL High quality results +/- 1 fps. In all resolutions...

No wonder I'm a little paranoid