[Wien] Large memory consumption of MPI k-point parallel version

Oleg Rubel rubel at Physik.Uni-Marburg.de
Thu Apr 10 15:58:35 CEST 2008


Hallo Robert,

On Thu, 10 Apr 2008, Robert Laskowski wrote:
> you should have about 2GB per process. Are you sure you are not running modified sources,
> or you run only one job per node. MPI runs are sometimes not cleaned after crash, they stay and occupy memory.

I would exclude this posibility.

> Your structure looks like surface, so why do you use 2 k-point in c direction.

I just play with MPI. I needed more k-points for the test.


Regards,

Oleg


More information about the Wien mailing list