Fillrate, GL_EXT_reme, Rightmark
Fillrate
Basic multi-texture fillrate scores give us an idea of efficiency when using the texture units, the score generated as a function of the GPU's clock speed. So with architectures as broadly similar as R480 and NV45, in terms of texture sampling units available, it's a decent basic comparison. It doesn't tell you anything about how fast common shaders will execute, or the latest games, rather just a figure of raw power. 3DMark05 provides a pair of good fillrate tests that aren't memory bandwidth bound.
Despite a 24% clock speed advantage, X850 XT PE is slightly slower than the 6800 Ultra when working with one texture. However, its inherent advantages when multi-texturing are shown off. Clock speed rules.
GL_EXT_reme
GL_EXT_reme uses simple OpenGL tests to stress the graphics card and find theoretical limits in a number of tests. It measures overdraw, basic texture cache efficiency when reusing a large texture, basic fillrate, geometry performance and hidden surface removal. In other words, most of the basic tasks a GPU performs to render 3D.
The 6800 Ultra is only able to come out on top when performance the texture cache effiency test (which stresses memory bandwidth), with the X850 XT PE storming ahead in the rest of the tests by at least the ratio of their core clock speeds.
Rightmark
D3D Rightmark, version 1.0.5.0 of which is able to benchmark Shader Model 3.0 cards using native shaders, measures much the same things as GL_EXT_reme.
Rightmark's HSR test favours the 6800 Ultra, with the fillrate tests also doing well on that card when the texture size increases, hinting that NV45 has better large texture performance than ATI's entrant. Note that as far as Rightmark is concerned, the PS2_a pixel shading test runs fine on X850, although it seems that a PS2_0 shader is being used in place during execution. X850 XT PE generally performs excellently, faster than 6800 Ultra in the basic fillrate, pixel shader and vertex shader tests.
