Re: Replica exchange on GPU with OPA network and Slurm

From: Geist, Norman (norman.geist_at_uni-greifswald.de)
Date: Mon May 11 2020 - 05:24:54 CDT

• Next message: M. A: "Re: FATAL ERROR: Charge placement error in QM-MM bond."
• Previous message: Thibaut Very: "Replica exchange on GPU with OPA network and Slurm"
• In reply to: Thibaut Very: "Replica exchange on GPU with OPA network and Slurm"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

There's no issue at all. Also best performance comes with
+devicesperreplica set

Bests
Norman Geist

Am Montag, den 11-05-2020 um 10:20 schrieb Thibaut Very:

Hello,

I would like to know if there is a way to configure NAMD and Charm++
to
run Replica exchange job with GPU support over several nodes on our
supercomputer?

We have a Intel Omni-Path network and our job scheduler is Slurm. Each

node has 4 V100 GPUs.

>From what I have seen on the mailing list, there are mutually
incompatible options to compile CHARM++ and NAMD to solve the
different
problems.

What our users would like, is to run with NVidia GPUs on several
nodes.

We run NAMD 2.13 and CHARM++ 6.10.1.

Best regards

-- 
Thibaut Very
Support Utilisateurs·trices
thibaut.very_at_idris.fr
+33169358569
http://www.idris.fr
IDRIS - Bat.506 - BP 167 - 91403 Orsay Cedex - France

• Next message: M. A: "Re: FATAL ERROR: Charge placement error in QM-MM bond."
• Previous message: Thibaut Very: "Replica exchange on GPU with OPA network and Slurm"
• In reply to: Thibaut Very: "Replica exchange on GPU with OPA network and Slurm"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2020 - 23:17:13 CST