[DFTB-Plus-User] performance on ryzen cpus

Ben Hourahine benjamin.hourahine at strath.ac.uk
Wed Dec 12 19:18:22 CET 2018


Hello Karol,

what are you using for BLAS operations? Reasonable size ground state
calculations use >95% of the cpu time inside LAPACK/BLAS.

Regards

Ben

On 12/12/18 17:53, Karol Strutynski wrote:
> Hello,
> My group have a machines with ryzen EPYC processors. I noticed that
> the performance of DFTB+ (version 18.2) is much worse than on
> equivalent intel cpu.
> I tried various compilers (gcc , ifort, aocc flang) with variation of flags.
> The best I got was using intel 2018 compiler with "-O3 -qopenmp -ip
> -standard-semantics -heap-arrays 10 -axAVX  -xHost -fast" flags.
> The results are usually the same - intel cpu is much faster than EPYC,
> in extreme cases the times are almost three times larger.
>
> The other programs compiled on the ryzen present performance similar
> to intel machine.
>
> It is quite possible I missed something during optimization. Does
> anybody have experience with running dftb+ on amd based machines?
>
> Best Regards,
> --
> Karol StrutyƄski
> CICECO - Aveiro Institute of Materials
> Department of Chemistry
> University of Aveiro
> Office 29.3.25
> 3810-193, Aveiro, PORTUGAL
> Phone: +48 739259476
> Mail: strutynski.karol at gmail.com, skarol at ua.pt
> ORCID: http://orcid.org/0000-0001-8733-9012
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at mailman.zfn.uni-bremen.de
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user

-- 
      Dr. B. Hourahine, SUPA, Department of Physics,
    University of Strathclyde, John Anderson Building,
            107 Rottenrow, Glasgow G4 0NG, UK.
    +44 141 548 2325, benjamin.hourahine at strath.ac.uk

2013/4 THE Awards Entrepreneurial University of the Year
      2012/13 THE Awards UK University of the Year

   The University of Strathclyde is a charitable body,
        registered in Scotland, number SC015263




More information about the DFTB-Plus-User mailing list