On Thu, Jul 16, 2015 at 6:18 PM, David A Case <david.case.rutgers.edu>
wrote:
> On Thu, Jul 16, 2015, Asmita Gupta wrote:
> >
> > I have some basic questions regarding running SGLD simulation in amber14.
> >
> >
> > The total no of atoms in my system are 359214 (including waters), but i
> am
> > printing only 1-864 atoms in the trajec file..However, a run of 1ns takes
> > approx 72-75 hrs to finish. (40 xeon cores , E52680)
>
> Which program are you using? How many nodes do you have, and what is the
> nature of the interconnect between nodes?
>
I am using pmemd.MPI (2 nodes, 20 cores each, IB card: Mellanox MT27500
Connectx-3).
> >
> > 1.) In explicit solvent conditions, does the guiding force is limited to
> > only solute or is applied to solvent molecules also? Is this why 1ns run
> is
> > taking this much time?
>
> SGLD is probably not the culprit here. You can easily compare timings for
> a
> short run with and without SGLD to compare timings.
>
I compared short test runs without SGLD, same script as above but with no
isgld, tsgavg, tempsg parameters. The timings have reduced considerably
without these (1.33 ns/day).
>
> >
> > 2.) Can i use something like isgsta=1 and isgend=864 to specify the range
> > on which guiding force should be applied?
>
> You can try this. I doubt that it will affect the timings, but you can
> easily
> enough experiment to find out. I'd suggest trying some test simulations
> with
> many fewer waters, to get a feeling for what is affecting your performance.
> Don't forget to try fewer cores as well.
>
> ...dac
>
Yeah, i tried on lesser cores as well. Although the timings are variable,
but in all the cases sgld runs are considerably more time consuming that
corresponding normal runs. So i feel that it is SGLD parameters which are
making this much difference. Can some adjustments be made in these to
reduce computational time without compromising sampling?
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Jul 16 2015 - 07:30:02 PDT
