Hi everyone,
So the Intel code has been in master for about a month now. We plan on releasing the patch within a week.
This is the current manual that will probably accompany it (maybe simplified): https://www.dropbox.com/s/e3hq1p494zwwz0t/Amber_OpenMP_Release_Documentation_v2.doc?dl=0
If anyone has any objections speak up now or forever hold your peace.
Charlie Lin
Graduate Student
Walker Lab
San Diego Supercomputer Center
________________________________________
From: Rai Nitin [rai.nitin.tcs.com]
Sent: Friday, May 20, 2016 6:08 AM
To: david.case.rutgers.edu; AMBER Developers Mailing List
Subject: Re: [AMBER-Developers] AMBER Master Branch
The new patch should mainly improve parallel performance.
For GB, one has to compile using MPI + OpenMP.
To build MPI + OpenMP
If you have already built an MPI version you can take a backup of
pmemd.MPI executable since the
MPI + OpenMP version is also named pmemd.MPI
cd $AMBERHOME
make clean
./configure ?intelmpi -openmp intel
make install
Running GB simulations:
export I_MPI_PIN_MODE=pm
export I_MPI_PIN_DOMAIN=auto
mpirun -np NSOCKETS \
-env OMP_NUM_THREADS=NCORES*2 \
-env KMP_AFFINITY="scatter,granularity=core" \
-env KMP_STACKSIZE=10M \
$AMBERHOME/bin/pmemdMPI ARGS
Running PME
export I_MPI_PIN_MODE=pm
export I_MPI_PIN_DOMAIN=auto
mpirun -np NSOCKETS * NCORES \
$AMBERHOME/bin/pmemd.hybrid.MPI ARGS
Substitute actual numbers for NSOCKETS and NCORES and arguments instead
of ARGS
Example:
If you have a 2 socket server, each with 10 cores
Then NSOCKETS = 2 and NCORES = 10
FOR GB
MPI ranks=2, number of openmp threads = 20
FOR PME
MPI ranks=20
Please let us know if you face any problems.
From: David A Case <david.case.rutgers.edu>
To: AMBER Developers Mailing List <amber-developers.ambermd.org>
Date: 20-05-2016 16:53
Subject: Re: [AMBER-Developers] AMBER Master Branch
On Mon, May 16, 2016, Charles Lin wrote:
>
> I was hoping if anyone had time they'd be able to help test out the
> current changes to pmemd in the master branch. We plan on pushing out
> this patch in around 2-2.5 weeks time. It has been sitting in the
> master branch since AMBER16 release. Its a fairly big update that
> increases performance around 1.2x-2x for CPU therefore it effects a lot
> of key files and routines in the code.
The new code compiles and passes all the tests on Intel 16.0.3 + MKL
under Linux.
Can you say where we should look for performance increases? My tests on
the jac benchmark for the serial code show no change in timings, compared
to
the amber16-with-patches branch. Haven't tried any other benchmarks yet.
(CPU is Intel(R) Xeon(R) CPU E5-2650 v3 . 2.30GHz.)
...dac
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
=====-----=====-----=====
Notice: The information contained in this e-mail
message and/or attachments to it may contain
confidential or privileged information. If you are
not the intended recipient, any dissemination, use,
review, distribution, printing or copying of the
information contained in this e-mail message
and/or attachments to it are strictly prohibited. If
you have received this communication in error,
please notify us by reply e-mail or telephone and
immediately and permanently delete the message
and any attachments. Thank you
_______________________________________________
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
Received on Thu Jun 02 2016 - 15:30:03 PDT
