Shadermark 2.1, RTHDRIBL
ShaderMark 2.1
Being able to benchmarking Shader Model 3.0 cards, Shadermark 2.1 is a decent test. It's run in both 2_b and 3_0 modes on the 6800 Ultra, for comparison to the X850 (2_b only). Compared to Shadermark 2.0, 2.1 is more friendly to NVIDIA hardware, the shaders used more able to take advantages of NV45's ability to do more GPU operations per clock cycle than R480.
The two tests that Shadermark 2.1 can't run on X850 involve blending floating point buffers, something R480 can't do. In the other shader tests, Shadermark 2.1 favours 6800 Ultra almost across the board. To interpret that, you simply understand that the shaders the program uses aren't massively general in nature, only testing single instructions or easy combinations of instructions that would execute in the same number of cycles on each.
The shaders are compiled for each GPU by their respective driver, before being executed. Therefore 'intended' workload, in terms of a shader program, and actual workload are somewhat different, due to the architecture of each GPU.
For the flipside...
RTHDRIBL
Shaders that favour ATI's architecture in play, in our real time HDR demo, using image-based lighting (projecting image data onto surfaces for use as light data).
