[QE-users] use of blas&lapack from amd-optimized libflame 2.1 causes test failures in epw tests

Tobias Kloeffel tobias.kloeffel at fau.de
Mon Feb 3 08:46:26 CET 2020


Recently I played around with some AMD EPYC cpus and the bad thing is 
that I also saw some strange numbers when using libflame/aocl 2.1.

The 'workaround' is probably to just use Intel's MKL. Since version 2020 
the MKL performs rather well when using AMD cpus, however, if you want 
to get the best performance you have to additionally set:

export MKL_DEBUG_CPU_TYPE=5

which gives an additional 10-20% speedup with MKL 2020, while for 
earlier versions the speedup is greater than 200%

Kind regards,
Tobias

On 2/2/20 10:57 PM, Paolo Giannozzi wrote:
> On Sun, Feb 2, 2020 at 10:23 PM Evgeny Permyakov <permeakra at gmail.com 
> <mailto:permeakra at gmail.com>> wrote:
>
>     sidenote: I had to set  **orte_execute_quiet = true** in
>     ~/.openmpi/mca-params.conf so mpi didn't mess with extra messages
>
>
> thank you for  this useful piece of information: I got rid of those 
> annoying messages
>
>     I don't use epw in my work and pwscf tests work just fine, but the
>     fact that some tests failed bother me. Should I investigate it
>     further and how? I'm not versed in QE code, I'm a humble user.
>
> investigating further would be a good thing, but it is not easy: one 
> has to figure out which piece of code fails, then why. Even if one 
> succeeds, very often the problem is in the library itself and  there 
> is nothing that can be done.
>
> Frequent offenders are complex-valued routines (e.g. zdotc) that seem 
> to follow two different incompatible conventions, but the symptom is a 
> hard crash, not a wrong number.
>
> Paolo
> -- 
> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>
>
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users


-- 
M.Sc. Tobias Klöffel
=======================================================
HPC (High Performance Computing) group
Erlangen Regional Computing Center(RRZE)
Friedrich-Alexander-Universität Erlangen-Nürnberg
Martensstr. 1
91058 Erlangen

Room: 1.133
Phone: +49 (0) 9131 / 85 - 20101

=======================================================

E-mail: tobias.kloeffel at fau.de

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200203/2b780eb9/attachment.html>


More information about the users mailing list