On Jun 22, 2015, at 1:01 PM, Anubhav Kumar <kanub...@iitk.ac.in> wrote: > Dear Filippo > > Thank you for your reply. >> Compile the code in serial, make sure you export CUDA_VISIBLE_DEVICES >> accordingly to target the GPU you want. > > Can't we run multiple mpi processes on single gpu?
Yes, you can but no more than two. QE try to maximize memory usage on the GPU so if you put more than 2 MPI per single GPU it becomes slow. In your case you do not want to use more than 4 MPI in total. 1:1 is still the best scenario. To better use the GPU in this way you need to use NVIDIA MPSS. Setting up NVIDIA MPSS is trivial but you need again to do the binding manually. This is not something that a non-expert user can do without support. I suggest to look for someone in your institution that is expert of GPU or a expert sys admin that can help you with this scenario. By email... it is kind of difficult (and it is _not_ the scope of this mailing-list) HTH Cheers, F -- Mr. Filippo SPIGA, M.Sc. http://fspiga.github.io ~ skype: filippo.spiga «Nobody will drive us out of Cantor's paradise.» ~ David Hilbert ***** Disclaimer: "Please note this message and any attachments are CONFIDENTIAL and may be privileged or otherwise protected from disclosure. The contents are not to be disclosed to anyone other than the addressee. Unauthorized recipients are requested to preserve this confidentiality and to advise the sender immediately of any error in transmission." _______________________________________________ Pw_forum mailing list Pw_forum@pwscf.org http://pwscf.org/mailman/listinfo/pw_forum
