OK, I'll get them to you ASAP. Unfortunately stampede appears to have gone
(or is going) offline. I'll try and reproduce the test failures on one of
our local clusters.
-Dan
On Thu, Mar 13, 2014 at 11:57 AM, Scott Le Grand <varelse2005.gmail.com>wrote:
> I need your raw mdout files for the failed tests. The comparisons on their
> own are mostly useless.
>
>
>
> On Thu, Mar 13, 2014 at 10:52 AM, Daniel Roe <daniel.r.roe.gmail.com>
> wrote:
>
> > The reason I initially used intel compilers is they are the default on
> > stampede (and other HPC centers as well). However, the issue exists with
> > GNU compilers as well, version 4.4.6, mvapich 1.9, cuda 5.0 (diffs and
> log
> > attached). The differences are similar (though not 100% exactly) to those
> > seen with the intel compilers.
> >
> > Let me know if you need any more information.
> >
> > -Dan
> >
> >
> > On Thu, Mar 13, 2014 at 10:39 AM, Scott Le Grand <varelse2005.gmail.com
> > >wrote:
> >
> > > Why do you guys bother with Intel's compilers for the CUDA edition? I
> > > can't even get my hands on them without paying the big bucks so there's
> > > zero incentive for me to debug Intel compiler issues other than saying
> > > don't use them. That said, if it's broken with gcc, then it's
> > interesting.
> > >
> > >
> > >
> > >
> > >
> > > On Thu, Mar 13, 2014 at 9:24 AM, Daniel Roe <daniel.r.roe.gmail.com>
> > > wrote:
> > >
> > > > Hi All,
> > > >
> > > > First, this time I promise I am using an up-to-date GIT tree:
> > > >
> > > > commit 8f81fff98e4d3095d0c66070eb3375f6a72708c0
> > > > Merge: 4de7ced 72c975d
> > > > Author: Pawel Janowski <pjanowsk.eden.rutgers.edu>
> > > > Date: Thu Mar 13 09:55:44 2014 -0400
> > > >
> > > > I am running on stampede (Tesla K20m) using intel compilers 13.1.0,
> > cuda
> > > > 5.0, and mvapich 1.9. For pmemd.cuda.MPI the PME tests go haywire
> > > (absolute
> > > > error as high as 1.31e+06!!). No segfaults though. Diffs and log
> > > attached.
> > > > Going back to commit d8024087a4d8c4c1e801192839df57d760bcadd2 (Wed
> Feb
> > 5
> > > > 22:17:48 2014 -0500) fixes the problems, although I have not
> > > systematically
> > > > gone back to see at what commit the code breaks. The problems happen
> > even
> > > > if I run on just 1 thread. GB seems OK, and pmemd.cuda seems OK as
> > well.
> > > >
> > > > Has anyone else seen issues like these? I will try using GNU
> compilers
> > > next
> > > > to see if the issue happens with them as well.
> > > >
> > > > -Dan
> > > >
> > > > --
> > > > -------------------------
> > > > Daniel R. Roe, PhD
> > > > Department of Medicinal Chemistry
> > > > University of Utah
> > > > 30 South 2000 East, Room 201
> > > > Salt Lake City, UT 84112-5820
> > > > http://home.chpc.utah.edu/~cheatham/
> > > > (801) 587-9652
> > > > (801) 585-6208 (Fax)
> > > >
> > > > _______________________________________________
> > > > AMBER-Developers mailing list
> > > > AMBER-Developers.ambermd.org
> > > > http://lists.ambermd.org/mailman/listinfo/amber-developers
> > > >
> > > >
> > > _______________________________________________
> > > AMBER-Developers mailing list
> > > AMBER-Developers.ambermd.org
> > > http://lists.ambermd.org/mailman/listinfo/amber-developers
> > >
> >
> >
> >
> > --
> > -------------------------
> > Daniel R. Roe, PhD
> > Department of Medicinal Chemistry
> > University of Utah
> > 30 South 2000 East, Room 201
> > Salt Lake City, UT 84112-5820
> > http://home.chpc.utah.edu/~cheatham/
> > (801) 587-9652
> > (801) 585-6208 (Fax)
> >
> > _______________________________________________
> > AMBER-Developers mailing list
> > AMBER-Developers.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber-developers
> >
> >
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
-------------------------
Daniel R. Roe, PhD
Department of Medicinal Chemistry
University of Utah
30 South 2000 East, Room 201
Salt Lake City, UT 84112-5820
http://home.chpc.utah.edu/~cheatham/
(801) 587-9652
(801) 585-6208 (Fax)
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Mar 13 2014 - 12:30:02 PDT
