UT2003 - Botmatch

Here we'll look at the Botmatch element of the UT2003 benchmark.

Again, we can see that at the lower resolutions the GeForce4 Ti is actually slightly faster than the GeForce FX - this could be down to memory latencies as the memory interface of the GeForce FX may be better tuned to take into account the latencies of DDR-II RAM at more bandwidth intensive situations.

Being primarily CPU limited, adding 8x AF in this test doesn’t account for a huge performance hit, but we can see that it does make it a little more fill-rate limited by 1280x960 and up, with a worst case performance hit of 19%. Utilising 4X FSAA shows that by 1280x960 it appears to be hitting a bandwidth limitation, in this instance creating a maximum performance hit of 39%. Using both 4X FSAA and 8x AF together creates a further impact, and unlike previous tests it is a cumulative effect, with a 56% performance deficit at the highest resolution in comparison to normal rendering.

Given that the higher quality filtering of 8x AF is taxing the texturing end of the pipeline, while 4X FSAA is fill-rate free but requires more bandwidth, the fact that the performance hit for adding both FSAA and AF is cumulative here is quite interesting, because it would appear that in this instance AF is taxing the bandwidth as well. It may be possible that the reason for this could be due to UT2003 requiring many textures layer, and they are not all being sufficiently cached, meaning that texture layers are swapping in and out of the cache as the layers are changing.