[gmx-users] Optimizing performance mac osx

[Mark Abraham]

On Fri, Oct 25, 2013 at 11:19 AM, Tiago Gomes <tiagogomes89 at gmail.com>wrote:

> Hi,
>
> I am relatively new to the gromacs environment and would like to optimize
> performance for my mac pro (osx 10.6.8)
> with 8 cores (16 in hyper-theading). I´ve read that one can use
> the g_tune_pme, i guess with np = 16. Don´t know if using mpirun or gromacs
> compiled mpi
> would be faster.


Won't be. GROMACS's default built-in thread-MPI is designed for this case.
Do check out
http://www.gromacs.org/Documentation/Acceleration_and_parallelization


> I guess sin gromacs 4,5 mpirun is deprecated and mdrun
> automatically distributes the workload through all the cores, i think.
> We also have a 40 core condor cluster; would setting it up there increase
> performance?


Using more than one node for the same simulation is not useful unless you
have a high-speed network, e.g. Infiniband. A network over which one would
use Condor would generally not be suitable.


> I think also that the scaling depends on the number of atoms.
> Any info on this?
>

Depends very much on the hardware, compiler, model physics and simulation
composition, also. You should aim for something at least around 500-800
atoms per physical core, though.

Mark