From: Almeida-Hernández, Yasser, Dr. (yasser.almeida-hernandez_at_uni-due.de)
Date: Fri Jan 21 2022 - 07:17:19 CST
I was referring to the container in https://urldefense.com/v3/__https://catalog.ngc.nvidia.com/orgs/hpc/containers/namd__;!!DZ3fjg!qVH7g3Y5qanGvz4iIcIjtuWWQ_EPjkh67xfegXoHkPW86YOVM6du8yQse6jZ1HmPOw$ (2.13-multinode).
In any case, surely the compilation in the cluster is always better. I just wanted to find a workaround for it.
Yasser
________________________________
From: Giacomo Fiorin <giacomo.fiorin_at_gmail.com>
Sent: Friday, January 21, 2022 1:51:11 PM
To: NAMD list; Almeida-Hernández, Yasser, Dr.
Subject: Re: namd-l: Running NAMD Singularity container on GPU multinode cluster with Omnipath
Is that the container you are referring to?
https://urldefense.com/v3/__https://singularityhub.github.io/singularity-hpc/r/nvcr.io-hpc-namd/__;!!DZ3fjg!qVH7g3Y5qanGvz4iIcIjtuWWQ_EPjkh67xfegXoHkPW86YOVM6du8yQse6jHaycrNg$
Because in that case, it quite clearly says "SingleNode" in the version label.
At any rate, for a cluster it is generally better to build the Charm++ library (which NAMD uses to achieve multi-node parallelism) to be better integrated with the cluster's software configuration.
Multi-node, however, will also require the older, hybrid CPU-GPU scheme and does not support the pure-GPU scheme of 3.0. With the latter you are limited to running one replica over one node at most, and perhaps do replica-based methods across multiple nodes (all is explained in the NAMD 3.0 alpha webpage).
Giacomo
On Fri, Jan 21, 2022 at 4:17 AM Almeida-Hernández, Yasser, Dr. <yasser.almeida-hernandez_at_uni-due.de<mailto:yasser.almeida-hernandez_at_uni-due.de>> wrote:
Dear all,
I want to run NAMD on several GPU nodes in a cluster with an Omnipath network. I thought to run the NVIDIA Singularity container. Does the container support the Omnipath network?
Best,
Yasser
This archive was generated by hypermail 2.1.6 : Tue Dec 13 2022 - 14:32:44 CST