I’m a relatively new CUDA programmer currently implementing GPU versions of some CPU algorithms that I’ve developed in the past. I’m working on an oct-core Intel i7 laptop with a humble Quadro K1000M (Fermi) GPU. Understandably, at the moment the CPU versions of my algorithms are beating the GPU versions by a mile. Althogh the plan is to run my algorithms on a high-end GPU on Amazon EC2, I can’t help but wonder which GPU would be at par with my CPU in order to do a meaningful comparison of both implementations. I mean if doesn’t make sense to compare the performance of an algorithm running on a sweet DGX-1 to an Intel Celeron, it shouldn’t the other way round.
Ideally, there would be a table where GPU and CPUs are ‘matched’ on some basis e.g. amount of silicon.