20 May
2017
20 May
'17
4:46 p.m.
On 17 May 2017 at 18:40, Octavian Lie octavian.lie@gmail.com wrote:
With FS6, -parallel -openmp x can really speed things up. I inherited code from prior versions using -usegpu flag for GPU/CUDA (not supported more recently). Is the latter redundant or affecting in any way the -parallel -openmp x performance, should we take it out, please advise.
I'm not aware of any specific problems - I compile with both CUDA and OpenMP enabled - but I am not an expert in bits of neuroanatomy which might be going awry. I can say that results obtained with the GPU will not be bit-for-bit identical to those using just the CPU. This is not unexpected, but might affect your decision on how to analyse your data.
Richard