[gmx-users] Several questions about the GPU version
mark.j.abraham at gmail.com
Tue Sep 3 17:23:44 CEST 2013
On Tue, Sep 3, 2013 at 9:37 AM, grita <cemilyigit at arcor.de> wrote:
> Hey guys,
> Meanwhile I run my simulations on a CUDA card and must say that I am very
> very satisfied with it.
> Nevertheless I have a few questions regarding to a improved performance:
> 1. Is it possible to perform the PME calculation on the CUDA card? I always
> see in md.log that the PME runs on the CPU, even if the flag -nb gpu is set.
I answered this last week. Was there something unsatisfactory about
that answer? :-)
> 2. Is it possible to run multiple simulations on one CUDA card? If so, how
> and is there an engine program for the management (like on clusters)?
Yes, but there is nothing fancy for it. See
You should just run multiple instances of mdrun, being sure to specify
a suitable number of threads so that your total usage fills all your
CPU cores. You will need to use the mdrun pinning options to pin
threads to cores manually, since mdrun will detect that it does not
have the CPU to itself, so it will not make unwarranted assumptions
about whether pinning is acceptable.
> 3. I don't understand the sense of the mdp option coulomb-modifier. What
> should it cause? I set it to none.
Have you checked out the appropriate part of manual 7.3?
> It would be very nice if someone or some people could answer the questions.
> Many thanks and best,
> View this message in context: http://gromacs.5086.x6.nabble.com/Several-questions-about-the-GPU-version-tp5010934.html
> Sent from the GROMACS Users Forum mailing list archive at Nabble.com.
> gmx-users mailing list gmx-users at gromacs.org
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
More information about the gromacs.org_gmx-users