ATi & nVidia in driver "optimizations" race

July 26, 2003 / by Leonidas / page 5 of 9

   ATi Optimizations

We choose a 16x anisotropic filter for these benchmarks in order to create a scenario which is limited by the graphics card for the reason to recognise driver optimizations more easily:

ATi 03.4 Pentium 4 Northwood 2.53 GHz  -  Radeon 9700 Pro 128MB 1024x768x32  -  16xAF   original driver   modified driver with AntiCheat script 3DMark2001 SE 11701 Pts. 11202 Pts. 3DMark03 (Build 330) 3937 Pts. 3869 Pts. AquaNox 2 69,5 fps 69,4 fps Codecreatures 26,1 fps 26,1 fps Comanche 4 42,3 fps 42,0 fps Counter-Strike 118,5 fps 117,4 fps Dungeon Siege 68,7 fps 68,6 fps Max Payne 65,1 fps 65,1 fps NOLF 2 63,8 fps 64,5 fps SSam: TSE 51,9 fps 51,5 fps Splinter Cell 36,7 fps 36,4 fps UT2003 (Flyby-Antalus) 121,5 fps 121,5 fps UT2003 (Pyramid2003) 43,7 fps 44,1 fps 20% 40% 60% 80% 100% 120%

This situation looks more tidy in addition to benchmarks by nVidia. By running both 3DMark benchmarks we have recognised minor optimisations used by ATi which can be identified clearly even if the effects are insignificant. It's impossible that the variation of +/- 10 points are measuring tolerances, because of the consistency that both 3DMarks show, even if there is a slight gap between measurements based on the original drivers on the one hand and the modified drivers on the other hand.

In these two cases it's the same as we mentiond at the benchmarks by nVidia: To use application specific optimizations along with theoretical benchmarks is absurd and should be called unallowed optimization. It's not the sense of theoretical benchmarks to score higher because of these specific optimizations. In this case there is no need to have a screenshot with lower image quality to underline our statement, because even the fact that application specific optimizations are used is enough to prove our point of view.

However, it's quite interesting in which way the newer versions of the ATi drivers behave concerning the two discovered "optimizations". In addition to that we tested both 3DMarks with newer driver versions called 03.5 and 03.6. In case of the 03.5 driver we made an extra test in terms of the modified driver version. The 03.6 driver refused to work with the AntiCheat script, so we have no other choice than accepting the standard benchmark score:

ATi 03.4, 03.5 & 03.6 Pentium 4 Northwood 2.53 GHz  -  Radeon 9700 Pro 128MB 1024x768x32  -  8xAF   original driver   modified driver with AntiCheat script 3DMark2001 SE (03.4) 11701 Pts. 11202 Pts. 3DMark2001 SE (03.5) 11729 Pts. 11212 Pts. 3DMark2001 SE (03.6) 11778 Pts. AntiCheat-Script nicht ausführbar 3DMark03 (03.4) 3937 Pts. 3869 Pts. 3DMark03 (03.5) 3885 Pts. 3878 Pts. 3DMark03 (03.6) 3884 Pts. AntiCheat-Script nicht ausführbar 20% 40% 60% 80% 100% 120%

These results dilute our statement mentioned before, since ATi already had announced, that there are no application specific optimizations for 3DMark03 within the 03.5 driver version. However current ATi driver versions still benefit from the application specific optimizations used for 3DMark2001 - including the driver version 03.5 and probably the newest driver version 03.6.