Hi Tommy, It really depends on the size of the simulation. Generally, in our experience, for larger real-world problems the compiler makes very little difference (4-5% maybe) with the rest coming down to BLAS/MPI. There are several posts on the mailing list archive describing the various BLAS and MPI options available along with my experiences of them. It is certainly worth reading through the archives!
The larger 3D problems -- of which there should be a few examples of on the mailing list -- are the ones you want to be investigating/drawing your attention towards. Regards, Freddie. CC: pyfr mailing list. ________________________________ From: Tommy Han [[email protected]] Sent: 30 May 2015 05:31 To: Witherden, Freddie Subject: Problem using PyFR Hello, I'm Tommy. I asked Vincent, Peter E about the question below: With datasets we have gotten, I just found out that icc runs faster than gcc in 2D datasets, while in 3D datasets it's just the opposite, gcc runs faster than icc. Is that true for all the datasets? There are too few datasets for us to check it out. Could you please tell us the result? Is it true for (almost) all datasets or it's just a coincidence? And below is what he answered, I have not seen this before myself. But I normally run with the CUDA backend. Freddie (cc’d) may be able to comment. Could you please answer it for me? -- You received this message because you are subscribed to the Google Groups "PyFR Mailing List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send an email to [email protected]. Visit this group at http://groups.google.com/group/pyfrmailinglist. For more options, visit https://groups.google.com/d/optout.
