Unreal Tournament 2003 is a popular game used to measure the performance of graphics cards under Microsoft's Direct3D API (Application Programming Interface). Using the outdoor-based DM-Antalus map for our testing grounds, the initial results are respectable for a budget graphics card.
Without antialiasing, the Botmatch results (red bar) indicate a CPU limitation as the drop in the average frame rate between the 640x480 (65 fps) and 800x600 (62 fps) resolutions is limited to 5 percent. On the other hand, frame rate variations in excess of 25% indicate that the graphics sub-system plays a greater role in determining performance than the CPU in the Flyby demo (green bar). These results are expected since the Flyby demo requires fewer CPU cycles, which will increase substantially in the Botmatch demo as a result of the processing associated with artificial intelligence and physics algorithms.
UT2003 Performance No Antialiasing / 2X Antialiasing
Looking at the Botmatch results and accounting for the decrease in performance by enabling sound, one could guesstimate that Soltek's SL-5200-XD will provide acceptable gameplay performance at resolutions of 640x800 and 800x600 when paired with a mid-range or better CPU.
Enabling 2X antialiasing at 640x480 caused performance to drop by 19% in Flyby (from 156 to 127 fps) and 9% in Botmatch (from 65 to 59 fps). The SL-5200-XD is in pretty good shape here and 2X antialiasing at 800x600 provided 50 fps in Botmatch.
In most cases we find that the GeForce FX 5200 doesn't have the necessary memory bandwidth to make 4X antialiasing practical at resolutions greater than 640x480. In Botmatch at 640x480, moving to 4X antialiasing resulted in a 26% decline in the average frame rate (from 65 to 48 fps) and a decline of 35% (from 65 to 42 fps) with 4XS. At 800x600 the decrease in performance is 45% moving to 4X antialiasing and 54% with 4XS.
Let's turn our attention to anisotropic filtering, which is an advanced filtering technique that improves the quality of textures at the expense of performance. A goal for this series of tests was determining the settings that provided at least 50 fps in the Botmatch benchmark with 2X antialiasing and all levels of anisotropic filtering enabled. This criterion should provide acceptable gameplay performance in UT2003.
UT2003 Performance Anisotropic Filtering
The 50 fps criteria could only be met at a resolution of 640x480. In CPU limited situations, the performance hit from anisotropic filtering is much less severe than that of antialiasing. Even with 2X antialiasing, the loss in performance due to anisotropic filtering at 2X, 4X, and 8X are 5, 10, and 14% respectively.