Whats the problem with 3DMark99 max?

[Stez2]

I recently noticed some weirdness with the 3DMark99Max scores on my rig (P2@560MHz, TNT2 32Mb, 128Mb blah blah)

As I overclock the card, the scores essentially do not change!

here's some results at 8x6x16:
(core/mem)

125/143 (default) 4805 pts
150/175 (oc'd) 4808 pts
170/175 (oc'd) 4807 pts

I thought it strange considering Q2 scores increased with vid speed accordingly..
Then I jump onto Ants Site, and see that he has seen the exact same phenomenon during his testing of various G400's with various CPU's....
So I decided to rifle off a mail to Ant to get his thoughts, in reply he feels that it is a driver limitation of some sort - sounds resonable to me. So I coninued testing at higher resolutions (should overcome driver limitations, yes?) and found that even at 16x12x32 the scores hardly changed at all (mind you they were pretty bad at that res - like 7fps or sumpthin hehe)..

Anyone care to speculate as to whats going on? My initial thoughts were that it must be a CPU limited benchmark, at least thats what the numbers suggest...

Obviously I can only test with my TNT right now, well until I get the guts to go buy a G400 that is..

here's some numbers from Matroxusers, compared with mine (this is G400 vs TNT2 as well)..

CPU | Score| CPU |Score
P2-560 | 4805 | +24% | +24%
P2-450 | 3861 | +25% | +21%
K63-400 | 3185 | Ref | Ref

Interesting yes?

James B.

[Joelg]

So what's your point? The score for 3DMark99Max is only based on the two game runs at the begining anyway. They take the two averages add them together and multiply by 50.

Joel

[Kruzin_K]

Yea...the final "xxx 3DMarks" don't mean that much. Look at the details of all the tests to get the full picture...

[Stez2]

I guess the point I'm trying to make here is that no matter how I clock up my vid card, the FPS (and therefore score) in 3DMark99Max remain the same. The FPS I see are the same or +/- 0.1 fps...
But in just about all (not every) game I play the difference when overclocked is noticeable to the eye, and during benchmarks.

The thing that bothers me is that this is touted as a video card benchmark, but these tests suggest it to be a CPU benchmark... The results support this theory across different CPU's and Video Cards..

Still I guess there's always a chance that BOTH the Matrox and Nvidia cards have massive driver limitations..



[This message has been edited by Stez2 (edited 08-08-99).]

[Lee KC]

Perhaps, the graphics card is "waiting" for the CPU, hence overclocking the card does not help much in the scores?

[M Buuri]

That's right. Something else is holding it back.

And I bet the games you're referring to (that will take advantage) are old, like Quake 2. That's because they have low polygon count, hence the graphics adapter will usually running on it's limits.

One should really keep in mind that 3DMark is NOT a graphic card benchmark, but a overall game performance benchmark. It's not designed to be CPU independent in any way. And that is why the score is relying to those 2 tests so much.
3DMark is a very good test and usually when somebody says otherwise he doesn't understand what it's for in a first place.

B


[This message has been edited by M Buuri (edited 08-08-99).]

[franksch3]

Hey Stez2,

here are some benchmarks I did with my G400:

3DMark99 Pro (not MAX).
Pentium II 450 MHz
128 MB
Matrox G400 32 MB SGRAM dualhead.

Default speed:
Resolution 3DMarks Game 1 - Race Game 2 - First person Fill rate Fillrate with mutitexturing
640x480x16 3540 58.5 59.5 169.9 189.9
640x480x32 3241 53.2 54.8 126.2 148.9
800x600x16 2994 50.3 49.5 176.0 195.9
800x600x32 2682 44.5 44.9 129.6 152.7
1024x768x16 2382 41.0 38.4 180.3 200.5
1024x768x32 2042 34.3 33.8 128.3 156.3
1280x1024x16 1781 31.1 28.3 180.7 201.6
1280x1024x32 1474 24.9 24.2 127.9 154.6

Overclocked to 300 PLL ie 150 MHz clock speed:
Resolution 3DMarks Game 1 - Race Game 2 - First person Fill rate Fillrate with mutitexturing
640x480x16
640x480x32
800x600x16
800x600x32
1024x768x16
1024x768x32 2307 38.6 38.3 156.7 186.0
1280x1024x16 2010 35.0 32.1 214.4 239.0
1280x1024x32 1708 28.9 28.0 156.7 187.4

Had no time yet to test the lower resolutions. But as you can see I get a pretty good speedup. I use 3DMark99 Pro (not MAX).

My Quake2 scores only got 4 extra fps at 1024x768x32, in demo1.dm2 because the OpenGL is the limiting factor.

Hope this might help. 3DMark99 seems to be quite processor limited.

Frank Schoondermark

[M Buuri]

Those results are not comparable to MAX-version results.

B

[Stez2]

Ok then, so looking at the tests done by Ant from Matroxusers and myself, and remembering that this is a GAME benchmark, I guess I can assume that a G400 will perform exactly the same as a TNT2?

lets recap here: (800x600x16)

P2-450 with G400 32Mb
3861 3DMarks

P2-560 with TNT2 32Mb (24% faster CPU)
4805 3Dmarks (24% higher score)

Hmmm.. sure seems like a test of CPU speed to me, but hey if the G400 works the same as my TNT2 then I'm in, hell it's got better TV out.. Anyone know if it runs under NT4 w/out hassles?

M Buuri:
I'm not sure what you call old vs new games, but here's what I've noticed when overclocking:

Quake 2 - noticeably faster
Quake 3 test - noticeably faster
Aliens Vs Predator - little faster
Sports Car GT - noticeably faster
Colin McRae Rally - no difference (but maxes at 800 res so can't really tell)
Grand Prix Legends - no difference
Midtown Madness - no difference
NFS 4 (high stakes) - little faster (10x7x32 only)

I think theres' a few recent games there, well they're the ones I play anyway.. oh and good old Quake 1 when u just want a fast frag fest