From: Vermaas, Joshua (Joshua.Vermaas_at_nrel.gov)
Date: Fri Oct 06 2017 - 10:43:59 CDT
Take a look at /home1/00288/tg455591/NAMD_scripts/runbatch_latest_knl. NAMD scales quite poorly if you use the default 64-68 processes per node. 13 (for whatever reason), behaves pretty well.
-Josh
On 10/06/2017 08:13 AM, Giacomo Fiorin wrote:
I don't have an account there. Since you're using it, you should check.
Not sure whether which username he goes by there (TACC uses their own system), but it may be possible to use:
getent passwd | grep Phillips
Giacomo
On Fri, Oct 6, 2017 at 9:32 AM, Chitrak Gupta <chgupta_at_mix.wvu.edu<mailto:chgupta_at_mix.wvu.edu>> wrote:
Hi Giacomo,
Thanks for the response. Yes, I did request 128 cores in the submission script.
Are you aware whether Jim has compiled versions and/or submission scripts for Stampede 2?
On Fri, Oct 6, 2017 at 8:50 AM, Giacomo Fiorin <giacomo.fiorin_at_gmail.com<mailto:giacomo.fiorin_at_gmail.com>> wrote:
Please find out how NAMD was compiled there and what part of Stampede you're using. Without this information, it's impossible to figure out your problem.
If the executable and the related environment module were provided by the TACC staff, you should contact them for clarifications.
Understand that the command line you pasted will launch two MPI tasks, each of which will spawn 64 threads. So you need 128 available cores on the node(s) that you requested.
For more information, look at the files notes.txt that comes with the NAMD distribution.
Giacomo
On Thu, Oct 5, 2017 at 7:23 PM, Chitrak Gupta <chgupta_at_mix.wvu.edu<mailto:chgupta_at_mix.wvu.edu>> wrote:
Hello,
I am trying to run NAMD on Stampede 2 but it is not scaling the way it should. Here is what I have in my submission script
module load intel/16.0.3
module load impi
module load namd/2.12
mpiexec -np 2 namd2 +ppn 64 filename.conf > logfile.log
I have also tried "ibrun" instead of mpiexec but the problem persists.
Best,
Chitrak.