Benchmarking Continued
Splinter Cell
Many people have the misconception that Splinter Cell is a DX9 game, or that it uses the absolute latest, most advanced pixel shaders which means it must be DX9. In truth, Splinter Cell uses an old codedrop (896) of the Unreal engine, and only requires DX8.1b to run. It's still a highly modified Unreal engine however, and the game itself uses many (older) pixel shaders to generate its many effects (i.e. night vision). The demo I ran, recorded in the level "The Caspian Oil Refinery," stresses fillrate and shading efficiency.
Well isn't this a surprise! Although Splinter Cell is DX8, the FX5600 Ultra doesn't seem to have that healthy lead over the R9600 Pro it did in Code Creatures, Aquamark, 3DMark01, and even UT2K3. Something to do with perhaps its pixel shader performance? Before we move on to the all crucial DX9 tests, lets take a very quick look at two OpenGL tests.
GLExcess
OpenGL support is really declining these days among consumer video manufacturers mainly because the latest OpenGL 2.0 isn't really for gamers and regular everyday users like us. I still like to run a couple OpenGL tests though for those of you still playing Quake or Counter-Strike.
The FX5600 Ultra has a slight lead at default settings and 2x8x AA/AF. When we go one step further however, its performance drops below the R9600 Pro drastically. Unexplainable.. really.
SPEC Viewperf 7.1
SPEC Viewperf's always been a must run benchmark to satisfy 3D modelers and such! AA and AF are disabled.
The FX5600 Ultra has a large lead in the dx-08 test while the R9600 Pro leads in the ugs-03 and drv-09 tests. For the rest of the tests, both cards perform pretty close. And alas we've reached the DX9 portion of this comparison.
3DMark03 Build 330
So the FX5600 Ultra generally outpaced the R9600 Pro in DX8 applications. But what about DX9? I'd like to quote from nVIDIA, the FX5600 is "designed for DirectX 9... guarantees powerful 3D experiences, with blazingly fast performance." Now it has been said that 3DMark03 isn't entirely DX9, but I think these scores are pretty self-explanatory.
How can the FX5600 Ultra "designed for DirectX 9" perform so significantly worse in comparison to the 9600 which the 5600 had consistently beat in DX8? Lets tack on AA and AF. Can the FX5600 Ultra pull ahead?
And the answer is a big fat no. Even with its superior memory bandwidth, the FX5600 Ultra cannot beat the R9600 Pro in 3DMark03 under any situation. Maybe this is just a fluke?
