ATi & nVidia in driver "optimizations" race

July 26, 2003 / by Leonidas / page 4 of 9

   nVidia Optimizations  (cont.)

The most important fact here is that nVidia uses pseudo trilinear filtering with a high portion of bilinear filtering. This wouldn’t cause much trouble, if it would be the standard mode for GeForceFX Chips. It gets problematic in comparison with other Direct3D applications, because nVidia uses strict trilinear filtering in the normal mode. The texture filtering behavior of Unreal Tournament 2003 with current nVidia drivers is an exception to this rule and this means that there must be some sort of application specific optimization. The following figures support this line of argument (from top to bottom: Direct3D AF-Tester, Devastation Demo, Serious Sam: The Second Encounter, 3DMark03):

Especially the second screenshot (taken from Devastation Demo) is of great interest. Devastation Demo uses the same engine like Unreal Tournament 2003 and most notably the same method of mipmap colorization but shows different texture filtering: Smooth transitions from one mipmap into another - proper trilinear filtering. This fact definitely shows that nvidia's driver 44.65 recognizes the execution of Unreal Tournament 2003 and alters the filtering technique from strict trilinear filtering to application specific pseudo trilinear filtering.

Finally it remains to be resolved to which drivers our conclusions apply to. We also took screenshots with driver versions 44.03, 44.67, 44.68, 44.71 and 44.90. They are virtually identical to the ones shot with version 44.65, so there's no use of publishing them here. We also verified the test for the reference filter quality with driver versions 44.03 and 44.90 which also delivered the same output. There is no visible rendered difference between these driver versions.

The only question that's left is, wheter or not the obtained results from the three benchmarks used - which receive application-specific optimization - result in measurable performance differences between individual driver versions. This would point out possible changes in already discovered application-specific optimizations. Therefore these three benchmarks with application specific optimizations, again with driver versions 44.03 and 44.90, respectively with and without AntiCheat script:

Judging from these results, it's safe to say, that all said before applies to all drivers from version 44.03 (current official driver) to 44.90 (latest unofficial version).