Unreal Tournament 2003 is a game with many new features and one feature we’re Epic very thankful for is the inclusion of a benchmark utility which resides in the /system directory, that makes it much easier for us to benchmark a 3D graphics card as we don’t have to write our own scripts or batch files anymore to execute benchmark runs in sequence. We’ve used the UT2003 benchmark as a means of determining the performance difference at various resolutions and with, or without, AA and AF enabled.
Both Nvidia’s and ATi’s drivers include various levels of anti-aliasing, AA, and anisotropic filtering, AF, which are user selectable and basically determine the image quality and performance. Now it would’ve been hard to compare these cards head to head if we weren’t able to find a setting that’s about comparable on both cards and what we can use for comparing the performance of both cards running with AA and AF enabled. After having received some feedback from both ATI and Nvidia we learned that ATI’s Quality and Performance mode are about similar and comparable to Nvidia’s Performance - Balanced setting, thus that’s what we’ll use for our benchmarks.
Take notice though that we’re talking about differences in image quality here, ATI’s Quality and Performance mode is comparable to Nvidia’s Performance - Balanced setting in terms of image quality. They both have a different approach in terms of hardware and we’ll see soon whether that has any drawbacks on performance. We used the 8x Performance - Balanced AF setting for the GeForce FX with 4 x AA and the 8x Performance mode for the Radeon 9700, again with 4 x AA, all other settings are left at default.
Upon looking at the above graphs, you can see that only in the lowest resolution the Radeon 9800 Pro and 9700 Pro manage to outpace the GeForce FX 5800 Ultra. In all other resolutions the GeForce FX 5800 Ultra reigns supreme. The gap is largest at 1600x1200x32 where the GeForce FX 5800 Ultra is 6% faster than the Radeon 9800 Pro and no less than 21% faster than the 9700 Pro. On the next page we’ll see whether this changes when we turn on AA and AF and really start to tax each card’s memory bandwidth.