[gmx-users] Thread, Multiple gpu, and sharing node

[Mark Abraham]

Hi,

CPUs and GPUs are not created equal, so there's not enough information to
say anything. But 3k atoms will not keep most GPUs fully busy, so you will
want to consider running multiple replicates of your system, e.g. with
mdrun -multi. See also http://arxiv.org/abs/1507.00898

Mark

On Tue, Aug 18, 2015 at 7:40 PM Johnny Lu <johnny.lu128 at gmail.com> wrote:

> Dear Users,
>
> Hi.
>
> The system has about 3k atoms, and I used a rhombic dodecahedron box.
>
> The node in the cluster has 2 sockets and 12 cores per socket. (24 sockets
> in total)
> It also has 4 gpu.
>
> Will I waste cpu time, if I run with all 24 cores and assign 1 gpu to 3
> cores?
>
> Should I compile for mpi?
>
> Alternatively, I can run with just 12 cores and assign 1 gpu to every 3
> cores.
> The remaining 12 cores and 2 gpu will be used by another user.
>
> Which way is a more efficient use of time per cpu core?
>
> Thank you.
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>