Variance of nvidia drivers in 3dmark

[zakelwe]

Hello all,

This is a continuation from the "optimisation " thread but I have given it a new thread as I have finished my testing.

Extremetech claim 43.45 and / or 44.03 drivers from nvidia unduly increase performance in GT2 because :-

"Clearing the buffers (color and Z) involves writing zeros to all memory locations in the new back buffer and Z-buffer so that the GPU can begin drawing the next scene with a "clean slate." But since not all memory locations will contain visible pixels, parts of the buffers remain dirty -- based on the assumption that the camera's unvarying movement path won't catch the errors. The advantage to not clearing buffers is the bandwidth savings of avoiding those writes to memory. This frees up precious memory bandwidth to be working on other things, like texture/texel reads and writes of visible parts of the scenes"

Here's my results for a GF4 ( no Gt4 testing as I cannot run that game).

First set comparing GT2 to GT3 over several drivers to see if GT2 has anomolies for 43.45 and 44.03 pushing them higher than GT3

Second set is investigating the known 42.68 issue where GT1 was not fully drawn.

You can get a feel of the nvidia drivers over time if you examine both sets. It makes a fun evening in on these cold winter nights.

I didn't cheat on any of them, they are all legit, and I am happy to answer any questions if they are unclear, like " what the fook does that prove ? "

http://www.aocb77.dsl.pipex.com/3dmark03.htm

Regards

Andy

[GlowStick]

Those tests you did are very good, but you cant test all of them unless you can get the developers versions of 3dmark03.......

I checked IRC, Kazaa and filemirrors. No dev version leaked up for grabs : O

[ragejg]

Originally posted by GlowStick
Those tests you did are very good, but you cant test all of them unless you can get the developers versions of 3dmark03.......

I checked IRC, Kazaa and filemirrors. No dev version leaked up for grabs : O

AAA!! He talky warez!

[zakelwe]

Originally posted by GlowStick
Those tests you did are very good, but you cant test all of them unless you can get the developers versions of 3dmark03.......

I checked IRC, Kazaa and filemirrors. No dev version leaked up for grabs : O

Hi Glowstick,

Yes I know. I was limited by not having an FX and not having the tool to see if the problem existed in which driver set for GT2 ( and GT4 if I had tested for the FX ) the bugs / cheats occurred.

The reason I did it was not to prove nVidia right or wrong, but i was very annoyed that Extremetech had all the tools and did no figures at all, simply said " A picture paints a thousand words ".

Lazy buggers LOL

I asked Dave Baumann whether 42.68 had the GT2 issue but has has not answered so I don't know.

I did find out some interesting stuff though from the figures :-

1) The Gt1 drawing problem was spotted with 42.68, but you can see that it actually occured at least as far back as 42.01 !Not only that but nVidia managed to increase the score from 42.01 to 42.68 massively ontop of the score got from the bug !

2) When the GT1 bug disappears it is not until much later, and the scores are still high. How ? Fair or foul ?

3) For GT2 high scores first become high at 42.68, not 43.35/44.03, so we need to know if the bug is in these, if so then nobody spotted it for a long time, ie beta testers not testing very efficiently ? If not then the bug/cheat does not increase performance.

4) 44.03 are crap for the GF4 and I think this might be a sign that nVidia is saying adios to coding for the GF4 range because the FX is so different. Sob sob.

Lets see how things progress though, I am sure it will all come out in the wash eventually.

Probably.

Possibly.

Kinda

Regards

Andy

[darkmiasma]

Originally posted by zakelwe
4) 44.03 are crap for the GF4 and I think this might be a sign that nVidia is saying adios to coding for the GF4 range because the FX is so different. Sob sob.

Andy -

The 44.03's on my system are awesome. I have no visual/rendering problems, and the IQ/Speed seem great. Plus the Fixed Aspect Ratio option for my LCD is awesome.

- Mike

[zakelwe]

I've now run some tests with build 320 and build 330 to compare for a GF4 in games GT1, GT2 and GT3

I used the previosuly fastest 42.68 drivers and the 43.51 drivers which were noted to have the bug in a similar fashion to 44.03

320 build

42.68
GT1 226.2 225.7
GT2 28.2 28.2
GT3 17.9 17.9
overall 3536 3533


330 build

42.68
GT1 227.4 226.5
GT2 19.6 19.7
GT3 18 18
overall 3232 3226


320 build

43.51
GT1 224.6 223.5
GT2 28.2 28
GT3 18 17.9
overall 3539 3510

330 build

43.51
GT1 213.7 211.4
GT2 19.6 19.7
GT3 18 18
overall 3130 3117


In build 320 the 42.68 drivers show the non drawing problem in Gt1 ( see 1) and 2) above in previous post ) , drivers 43.51 do not show this non drawing issue but still stay high .

In build 330 there is still the drawing bug in Gt1 in 42.68 but now when that is got rid of in 43.51 the scores drop, showing as well as the drawing bug there was another cheat /optimisation.

Note that in both sets of drivers in 330 Gt2 takes a big hit .

GT3 is bloody unimpeachable ... fantastically even results from 40.52 upwards. Is this test nvidia proof ?

Regards

Andy

PS Forgot to add, these are in 640x480 to try and show the differences more in GT2/Gt3 .. my GF4 is too slow otherwise.