On Tue, Apr 04, 2017 at 09:37:14PM -0400, David Case wrote:
> On Tue, Apr 04, 2017, Dan Roe wrote:
>
> > On Mon, Apr 3, 2017 at 11:26 AM, David Case <david.case.rutgers.edu> wrote:
> > > a. do the pmemd tests pass with PGI compilers?
> >
> > PGI 17.1:
> >
> > Finished serial test suite for Amber 17 at Tue Apr 4 09:51:06 MDT 2017.
> >
> > 162 file comparisons passed
> > 0 file comparisons failed
> > 0 tests experienced errors
>
> Thanks...that indicates that there might(?) be some speed advantage by
> using PGI vs. gnu for the cpu version of pmemd. Does anyone know if
> this is true or not?
Yes, i find bigger than expected gains for intel and pgi over gnu
on xeon e5-2680 broadwell [gpu is P100].
Resources Performance(s) ns/day exe
owens gnu nodes=1:ppn=28:gpus=1 415 208 pmemd.cuda
owens intel nodes=1:ppn=28:gpus=1 416 208 pmemd.cuda
owens gnu nodes=1:ppn=28 serial 44953 1.92 pmemd
owens intel nodes=1:ppn=28 serial 37936 2.28 pmemd
owens pgi nodes=1:ppn=28 serial 40887 2.11 pmemd
gnu/6.3.0
intel/16.0.3
pgi/17.3.0
https://www.osc.edu/resources/technical_support/supercomputers/owens/
GPU_Validation_Test.
Short is DHFR NPT MD: 23558 atoms, explicit TIP3P solvent,
particle mesh Ewald, 8 Angstroms direct space cut off, shake, 2 fs time
step, Berendsen 300 K, 500000 steps of isothermal-isobaric ensemble
molecular dynamics.
scott
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Apr 08 2017 - 17:30:03 PDT
