ATi & nVidia in driver "optimizations" race

July 26, 2003 / by Leonidas / page 2 of 9

   nVidia Optimizations  (cont.)

Because of that we can go on without image quality comparision in 3DMark, since only the performance discrepancy is sufficient here as proof for an unallowed optimization. In fact this does not include Unreal Tournament 2003: In a real game these application specific optimizations are legal and welcome, if they boost the performance without loss of image quality! Such optimizations are unallowed if the result is a loss of image quality.

To find out the reasons for the higher performance of the original driver compared to the unoptimized (modified) drivers in Unreal Tournament 2003, we had to dig a little deeper. At first we had to find out if this performance difference appears only in special flybys, botmatches or timedemos, or if they do in the whole game. To do so, we testet all the flybys, botmatches and timedemos available, at first without anisotropic filtering:

nVidia 44.65 @ UT2003 Pentium 4 Northwood 2.53 GHz  -  GeForceFX 5800 Ultra 128MB 1024x768x32  -  without AF   original driver   modified driver with AntiCheat script Flyby Antalus 149,2 fps 142,3 fps Flyby Asbestos 199,2 fps 198,3 fps Flyby Citadel 101,6 fps 102,8 fps Flyby Face3 145,4 fps 146,7 fps Flyby Inferno 176,5 fps 127,5 fps Flyby Phobos2 157,3 fps 150,4 fps Flyby Suntemple 177,8 fps 161,4 fps Botmatch Antalus 43,5 fps 44,1 fps Botmatch Anubis 59,0 fps 60,1 fps Botmatch Asbestos 57,0 fps 57,9 fps Botmatch Citadel 41,3 fps 41,7 fps Botmatch Inferno 64,0 fps 64,2 fps Botmatch Suntemple 57,0 fps 57,3 fps Timedemo Pyramid2003 44,8 fps 44,6 fps Timedemo Overkill 37,5 fps 34,9 fps Timedemo PCGH 73,0 fps 74,7 fps Timedemo Cheat 75,1 fps 77,9 fps 20% 40% 60% 80% 100% 120%

... and the same with 8x anisotropic filtering:

nVidia 44.65 @ UT2003 Pentium 4 Northwood 2.53 GHz  -  GeForceFX 5800 Ultra 128MB 1024x768x32  -  8xAF   original driver   modified driver with AntiCheat script Flyby Antalus 140,1 fps 77,1 fps Flyby Asbestos 175,2 fps 113,9 fps Flyby Citadel 100,7 fps 82,4 fps Flyby Face3 141,5 fps 106,0 fps Flyby Inferno 128,1 fps 64,4 fps Flyby Phobos2 143,6 fps 95,0 fps Flyby Suntemple 133,1 fps 85,1 fps Botmatch Antalus 43,6 fps 43,7 fps Botmatch Anubis 58,7 fps 60,4 fps Botmatch Asbestos 56,5 fps 58,2 fps Botmatch Citadel 41,0 fps 41,7 fps Botmatch Inferno 63,4 fps 39,3 fps Botmatch Suntemple 57,0 fps 55,4 fps Timedemo Pyramid2003 44,5 fps 41,7 fps Timedemo Overkill 34,1 fps 25,4 fps Timedemo PCGH 72,0 fps 65,8 fps Timedemo Cheat 74,9 fps 63,7 fps 20% 40% 60% 80% 100% 120%

Out of these tests it can be concluded, that nVidia has - despite of one or another outlier - probably built a general optimization for Unreal Tournament 2003 and no optimization only for certain flybys or only for the anisotropic filter. Those who think they can counter-cheat out these optimizations by using so called "custom timedemos" are wrong. Despite the fact, that the used timedemos "Overkill" and "Cheat" are unknown by nVidia, they worked with the optimisations. So we conclude, that the optimisations work in the whole game.