Quote Originally Posted by Calmatory View Post
As a programmer who has some serious interest at writing (CPU) benchmarks, I can only say that code A with compiler X on platform Y will just be faster on vendor Z CPU. This without ANY bias towards CPU vendor. A good example would be when I wrote sorting benchmarks; mt Celeron M @ 1.6 GHz was WAY faster at Bubble sort, Insertion sort and Selection sort than my brothers A64(Windsor) @ 2.4 GHz. Quicksort was actually slower on my Celeron. I bet that you would blame me for being Intel biased with the code, but no. Just generic C with GCC 4.2.0(Mingw for Windows, A64).

Though, I didn't check the compiler generated code to verify that the code was exact same(E.g. no SSE2 optimizations).

What I mean, is that two pieces of code can run with BIG performance differences, the benchmark author isn't biased when writing such code, IF the code between CPU vendors is identical. However, if in real world the executables are vendor-optimized, should the benchmarks be too?

Though, I'd agree with the statement that there should be some kind of a standard of CPU benchmarking. But what kind of? It will be somewhat biased(according to people) to either CPU vendor due to the nature of CPU.


Honestly, what the F***?!
3d rendering is quite a powerfull benchmark. Run a Mental Ray scene in max/maya for example and you will quickly see exactly what the difference is between different CPUs. It scales perfectly with as many cores/threads you have.