Why do AMD GPUs take 1 CPU core?

[iceman1992]

Okay I wasn't sure where to post this, but since this isn't about a specific hardware..
The first GPU folding was done on AMDs, but why did Nvidia overtake it in terms of performance?
I don't know about the newer generation of cards but I heard Nvidia was much better.
And of course AMDs use 1 full core whereas Nvidias barely use any.
Is it the driver? Is it the architecture or something?
Feel free to refute my statements, since I haven't used AMD for a few years, and have only folded with Nvidia.

[7im]

Yes drivers and architecture. Nv drivers are faster but less stable. Amd driver are very stable. Also has to do with NV having very mature CUDA software while AMD scrapped Brook and started over with OpenCL.

[iceman1992]

How do HD7000 and Nvidia 600 compare? Would you say OpenCL is a good (or better) alternative to CUDA?
Because my GPU is aging and sooner or later I will have to decide.

[7im]

I won't compare products that were just released, and that the fahcores have not been optimized for either. The performance on each type will change drastically as some point, but to what degree is unknown. Any guess now would be more damaging than helpful.

[bruce]

How do HD7000 and Nvidia 600 compare? Would you say OpenCL is a good (or better) alternative to CUDA?
Because my GPU is aging and sooner or later I will have to decide.

Today there are three possibilities: Most NVidia projects run on a CUDA based core with NV drivers. A few NVidia projects run on an OpenCL based core with NV drivers. All AMD GPU projects run on an OpenCL based core with AMD drivers. Comparing option 1 with option 3 is favorable to NV.

Unless you actually have data comparing option 2 to options 1 and 3 you can't really make a general statement about whether the differences are due to OpenCL or to driver differences.

The amount of CPU used by CUDA is quite small for simple proteins on relatively slow GPUs. As the assignments progress toward more complex proteins on faster GPUs, the amount of CPU used by CUDA increase and at some unknown point might reach the same level as AMD for the same proteins at similar speed. At this point it would only be guessing, and as 7im suggests, we'd have to also guess about what will happen after the FAHCore for Kepler (and beyond) is optimized.

[iceman1992]

Ah I see. I'll wait for the optimized core then. I really do hope AMD and NV will be equal later on, gives me a much wider choice
Thanks 7im and bruce