[Pw_forum] about MPI and OpenMP threads for QE-GPU
spiga.filippo at gmail.com
Tue Sep 1 15:54:09 CEST 2015
On Aug 25, 2015, at 1:46 AM, Mutlu COLAKOGULLARI <mutlucolakogullari at trakya.edu.tr> wrote:
> So, I am in trouble for sharing the cores between MPI and threads for OpenMP.
Simple rules valid for QE-GPU runs:
- one MPI for each GPU
- if number of GPU is equal number of socket in the node, set as many OpenMP threads as core per socket
- if number of GPU is not equal number of socket in the node, set OpenMP threads to fill all cores of the node or (probably better) do not use OpenMP at all (= MPI + GPU, --disable-openmp)
If you make it works following these rules and you get numbers out of it then you can double the number of MPI for each GPU (instead of 1:1 ratio, 2:1 ratio).
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."
More information about the users