The CDRS benchmark that everyone uses for bragging contests because the numbers are large is geometry-bound only on machines that do fast anti-aliased lines.
The dataset is small, and has line segments that map to less than one pixel. No image quality is mandated. The combination of these aspects allow people to tweak for it much too much to my liking. As with other benchmarks, its age means that people have too much chance to focus on it when designing hardware and software.
Over 60, it's pretty meaningless -- if you have a fast card, you'd rather show more complex models at 60fps, or the same model at higher quality, than show the same vacuum cleaner at 200 frames per second.
Still, disregarding its shortcomings, very high numbers do point out to a pipeline that handles geometry info very rapidly for problems that don't do anything very complex with OpenGL (i.e. CAD/CAM).
It's going to be replaced by ProCDRS shortly (the OPC council has developed ProCDRS exactly for the reasons detailed above), which uses a better mix of rendering modes, and a much larger model.
Unfortunately, not everyone has published ProCDRS.
And also, for bragging, CDRS will continue to be used for some time just because the numbers are large, and because companies will break the 'magical' 200 barrier, and these magical numbers look great on press releases.
CDRS numbers will be used even in markets that couldn't care less about CDRS because most applications in these markets will be fill-limited, and none will use 3D lines (witness IBM's latest gfx launch). |
