[DFTB-Plus-User] Running DFTB+ on Ibm-multicores
Bálint Aradi
balint.aradi at bccms.uni-bremen.de
Wed Apr 21 14:43:06 CEST 2010
Dear Giovanni La Penna,
> 1) Is the "time" command the proper way to measure
> performance for OpenMP applications? Are there other
> possibilities?
Yes, with the time command you can get a good guess for the performance
of your code, especially when comparing wall time (real time) vs. user
time.
>
> 2) Before trying to install a (maybe) multi-threading
> version of Atlas (that with "pt"), do you expect any improvement
> compared to the thread-safe atlas version (that without "pt",
> the only one installed on the system so far)?
For big systems DFTB+ spends about 95% of the time in the
diagonalisation routines, which are provided by the external library you
use. If this library is not OpenMP-parallelised, then you won't see any
improvement compared to the serial execution. So, you definitely should
use a threaded BLAS or LAPACK.
Threaded ATLAS would a natural choice for BLAS/LAPACK with threads. I
had though some difficulties with recent versions of it, but maybe the
current one works fine.
Best regards
Bálint
--
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
http://www.bccms.uni-bremen.de/en/~b_aradi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 252 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20100421/ea5a5849/attachment.sig>
More information about the DFTB-Plus-User
mailing list