Re: Far Cry Benchmarks - Shader Model 3.0 performance
As a developer I like SM3 but I have to say that you should not expect the same performance difference between SM2 and SM3 with each game that use SM3.
The new Farcry patch use two different technics to improve performance:
1. object instancing: Draws many of the same objects with different per instance data (like position) in one call. This helps to relieved the CPU. You can do this only with SM3. But if you have many small static objects like grass you can merge it to a meadow. In this case you do not need object instancing to get the performace.
2. reduce light passes: Far Cry use one pass per light. In the case you use SM3 it can merge up to 4 lights in one single pass. This means less calls (good for the CPU). Because there is some work that is equal for each light source this work will only done once for up to 4 source. This save Fillrate and Shaderpower (good for the GPU). You do not need SM3 for such things you only need a chip that can execute long shaderprogramms like every GeForceFX or the new Radeon X800. Even with the much smaller maximum programsize of the R3XX Chips you can merge some lights together. In most case you will not be able to merge 4 lights in one pass at the R3XX but 2 passes with 2 lights are better than 4 passes with one light.
What I try to say is that in the case of Far Cry you compare a to some extent optimized SM3 path with a bad optimized SM2 path.