ATi & nVidia in driver "optimizations" race

July 26, 2003 / by Leonidas / page 7 of 9

   ATi Optimizations  (cont.)

Fortunately the ATi driver leaves the user a small back door, if a genuine trilinear filter is preferred. Because when we enabled the anisotropic filter not via ATis control panel but used the tweak tools aTuner 1.4.32.4465 (comes with a experimental but in our benchmarks nevertheless efficient Radeon support) or the rTool 0.9.9.6d (the previous version 0.9.9.6c wasn´t able to do that!), we had no problems to unite the trilinear and the anisotropic filter (compare with the original ATi screenshot on the right):

At this place, there was of course the question, whether there are other differences between those two pictures aside of the bilinear or trilinear filter. After detailed investigation of appropriate "normal" screenshots (i.e. without colored MipMaps) we could however detect only one single minor difference (MouseOver effect with Javascript, alternative: click opens both screenshots):

If one looks at the edge to the abyss quite at the right edge of the abyss, then minimum differences can be seen in favor of the rTool. Since all other screenshots recorded by us including those with colored MipMaps however showed no further differences, one can roughly say that there is probably no difference in image quality between the bilinear/trilinear anisotropic filtering by the control panel and the pure trilinear anisotropic filtering by rTool, with regard to static elements. Excluded from that is the effect of mip-banding, which can only be seen in motion.

By the way there was finally a possibility at our disposal to control, how much computing power ATi saves with the bilinear/trilinear anisotropic filtering by control panel compared to the pure trilinear filtering by rTool. For this purpose we simply ran some benchmarks with the 16x anisotropic filter. The anisotropic filter was once regularly produced by the control panel, which results in a bilinear/trilinear anisotropic filter and once over rTool 0.9.9.6d, which results in a trilinear anisotropic filter.

We could observe by the way, that during these benchmarks the settings of the control panel and aTuner resp. rTool interfered with one another. In each case just the settings of the program run first were accepted and only a restart of windows solved this problem. First we ran some Unreal Tournament 2003 benchmarks:

Thus advantages gained by ATi's optimization of anisotropic filtering can at least be quantified for Unreal Tournament 2003. In the botmatches and also in the timedemos (replays of real gaming) changes can hardly be seen, which results from the general CPU limitation of the game. The fact that the general optimization of the anisotropic filter works nevertheless is noticeable in the flyby benchmarks which - free from CPU limitations - shows about 20 percent difference between bilinear/trilinear anisotropic filter (control panel) and pure trilinear anisotropic filter (rTool).