[DFTB-Plus-User] Problem of parallel computing with openmpi
aradi at uni-bremen.de
Fri May 12 08:08:05 CEST 2017
> Actually I use a normal 1 Gb networking switch, rather than an
> infiniband switch, so it may be one of possible problems.
> I will try a larger number of atoms varying cores and nodes.
That is usually not that well suited for parallelized diagonalization,
as diagonalization is rather communication intense and needs a low
latency interconnect. However, as Ben suggested, you can parallelize
over k-points, as the diagonalization in each k-point can be carried out
independently on the nodes (no communication needed).
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 836 bytes
Desc: OpenPGP digital signature
More information about the DFTB-Plus-User