Benchmark Test Scores
SimHQ included Futuremark’s 3DMark06 for examining the theoretical raster and shader performance characteristics of the two graphics boards. The individual fill rate (multi-texturing) and pixel and vertex (complex) shader tests were ran at the program’s default resolution of 1280×1024.
X1800 XT
|
X1900 XTX
|
|
Fill Rate |
9910 MT/s
|
10330 MT/s
|
Pixel Shader |
161 fps
|
268 fps
|
Vertex Shader |
56 MV/s
|
62 MV/s
|
Considering the small clock speed difference between the two cores the fill rate score similarity isn’t too surprising; the tripling of shader cores obviously has no impact on the X1900s raster performance in this particular synthetic test. The vertex shader test likewise is fairly similar between the two test boards, with the X1900 pulling roughly 10% ahead. The pixel shader test, however, is where the X1900s design is allowed to shine, with the tripled shader cores producing a score that is 66% faster than that produced by the X1800.
We also ran the two game tests inlcuded in the benchmark suite to see whether or not the X1900’s tripling of shader cores would allow it to pull substantially ahead of the X1800.
X1800 XT
|
X1900 XTX
|
|
Canyon Flight |
17.5 fps
|
25 fps
|
Deep Freeze |
17.5 fps
|
26.9 fps
|
Both game tests make fairly significant use of shaders, so the X1800 is again considerably outperformed by its refresh, with the 1900 pulling 50% ahead in both tests. As with all synthetic testing, care should be taken before extrapolating too many assumptions concerning real-world performance, but the performance scaling exhibited by the 1900 here is clearly due to the increased number of shader units; thus it shouldn’t be too bold to posit that future titles that make increasing use of shader instructions will likewise scale in performance.
Lock On: Flaming Cliffs was tested using the “SU-25T vs. two A-10” replay with the graphics settings at high (heat blrs off).
Heavily CPU-dependent, Flaming Cliffs shows little variance between the two test boards. The X1900 eeked out out the smallest of victories at the lower resolutions, though a 1 fps difference is obviously well within test error margins.
High quality (4x AA and 8x AF) shows the the same pattern, with the 1900 giving essentially identical performance, though there is more of a drop-off across the tested resolutions with these settings. The lack of scores at 1920×1200 is due to the fact that Flaming Cliffs would show no AA as being in use at this resolution.