ShaderMark 2.0
This should be the last time you see ShaderMark 2.0 make an appearance in the review of a reference board product at HEXUS, as I move to 2.1 in the future. Broken shader tests will be no more.
The AGP board is either as fast, or faster, than the PCI Express version using 66.93. The 1000MHz clocked version increases performance usefully in the shader tests that benefit from extra memory bandwidth (a surprisingly large amount). While it's feasible that the core logic used is exerting an influence over things, in this 99.99% GPU-bound test it's the driver controlling the difference. Using 66.93 on the PCI Express version shows virtually identical performance. It's clear that the bridge chip isn't affecting performance.
Compared to the Radeon boards, the 6600 GTs are heavier hitters across the board, either by a small amount or by up to 50% more. The basic architecture of NV43 and its efficient fragment shader performance is the reason why.
