[gmx-users] Free energy calculations and speed problems

[van Bemmelen]

Hi Justin,

You shouldn't use PME in vacuo. Just use pbc=no and infinite (i.o.w.: no) cutoffs.

I'm not entirely sure why PME takes such a long time, but I can give a very wild guess. It may have something to do with the fact that, if I remember correctly, the Fourier transform of one (or a few) large charge spike(s) can only be represented to a certain precision by taking very high frequencies into account. Therefore the FFT has to calculate ridiculously long to reach the desired accuracy.

Hope it helps,
Jeroen



Date: Mon, 19 May 2008 15:04:51 -0400
From: "Justin A. Lemkul" <jalemkul at vt.edu>
Subject: [gmx-users] Free energy calculations and speed problems
To: gmx-users at gromacs.org
Message-ID: <1211223891.4831cf53f0a46 at webmail.vt.edu>
Content-Type: text/plain; charset=ISO-8859-1


Hi all,

I'm running some free energy calculations (in vacuo and in solution) for a set
of molecules I am trying to parameterize.  All the molecules are fairly small,
about 30-35 atoms in each.  The solvated calculations go smoothly, and rather
quickly (~5 hr/ns on two nodes).  I am running into a problem when I do the in
vacuo calculations.  Since I am constraining all bonds, I am running things on
a local machine in the lab (single core AMD64 Opteron).  I get as far as NVT
equilibration (removing charges only) before I give up and kill it, since it
takes so long.  A 10-ps NVT run takes about 20 minutes, but that adds up to
over 5 days for my 5-ns production run!  The machine is no slouch; I have used
it to run other small NVT/NPT/MD runs with good performance.  I get similar
performance on another machine that is even faster (a single CPU of the 2-CPU
node on our supercomputer).  The installations of Gromacs on both are fine, as
far as I can tell, and both have been used extensively.

I am using Gromacs version 3.3.3, and I have attached my .mdp file and the tail
end of my log file for reference.  If anyone notices anything obvious that I'm
doing wrong, I would be grateful if you pointed it out!  I see in the log of
the slow run (in vacuo) that the 3D-FFT calculation is dominating the CPU
usage.  Is that indicative of any problem?

Thanks for your attention.

-Justin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20080520/b7769c1d/attachment.html>