Re: namd mpi

From: Brian Radak (brian.radak_at_gmail.com)
Date: Fri Jun 22 2018 - 08:20:16 CDT

In my experience, within a single node, multithreaded and MPI performance
are virtually identical. The only difference is that MPI has a larger RAM
footprint. Although I doubt this is a problem with any modern system with
8-16GB RAM, multithreading is thus probably preferable in all instances.

Pre-compiled multithreading+CUDA binaries are available for download in
order to utilize GPUs and this is definitely recommended unless you want to
spend a day futzing with compilers. I would not recommend running on GPUs
across nodes unless you have a cluster specifically built for this purpose
(i.e. with expensive interconnect).

HTH,
BKR

On Fri, Jun 22, 2018 at 8:46 AM, Stefano Guglielmo <
stefano.guglielmo_at_unito.it> wrote:

> Dear namd users,
>
> I have a kind of basic question: in case of a single node with more than
> one cpu (actualy two in my case) would NAMD benefit of compilation with
> OpenMPI? And if so, would it be possible for it to use gpu acceleration as
> well?
>
> Thanks
> Stefano
>
> --
> Stefano GUGLIELMO PhD
> Assistant Professor of Medicinal Chemistry
> Department of Drug Science and Technology
> Via P. Giuria 9
> 10125 Turin, ITALY
> ph. +39 (0)11 6707178
>
>

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2019 - 23:20:01 CST