[Pw_forum] OpenMPI or not?

Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu
Thu May 7 12:42:48 CEST 2009


On Thu, 2009-05-07 at 17:19 +0700, Vit wrote:
> Hi,
> 
> I've got a PC with Linux and 1 Phenom II x4 processor (4 core cpu). Do I 
> need OpenMPI to run 1 job in 4 threads? or I just can complile QE with 
> AMD acml_mp library and use the parallel calculations?

you can run with the multi-threaded BLAS, but that will only give you
some limited speedup. the MPI parallelization is much more efficient
in almost all, except a few corner cases. we had a discussion on that
subject here a while ago. the tricky thing is to find out, which of 
the different levels of parallelism, that Q-E offers is the most
efficient for your calculations. it used to be that the parallelization
across k-points used to be the most efficient one, but that has a 
larger memory footprint and will thus have a lower cache efficiency,
particularly on AMD cpus. we had a long discussion on that subject here
not so long ago with some benchmark tests from several people. please
check the mailing list archives and see, if there is anything useful
in it for you.

cheers,
   axel.
> 
> Koroteev Victor,
> Nikolaev Institute of Inorganic Chemistry SB RAS
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.



More information about the Pw_forum mailing list