0

I was reading this question about FLOPS and I agree with all that, but then how should we measure the performance of a program that does not contain floats? Should we go for MIPS (million instructions per seconds? If so, how to count them? Can someone give a small code and explain how to count MIPS?

Community
  • 1
  • 1
BRabbit27
  • 5,565
  • 15
  • 81
  • 150
  • 1
    Could you clarify the need to use CUDA and OpenCL tags? – BenC May 22 '13 at 08:15
  • Because I am interested in GPGPU performance more than in CPU, although throughput is throughput no matter what type of processor. Should I remove them then? – BRabbit27 May 22 '13 at 09:35
  • 1
    Since the same question could be applied to any computing hardware, I think that these tags are not needed, unless you have some CUDA-specific question to ask. – BenC May 22 '13 at 09:39

0 Answers0