[gmx-users] Performance

[Anton Feenstra]

M.Naser wrote:

> Hi All,
> 
> Is there any trick for improving the performance of a lam/mpi parallel run?

No, except waiting for the next version which should have a completely 
overhauled parallelization scheme.

> I am aware of the flags : sort. shuffle and load.

You write full precision out coordinates, velocities and forces 
(nstxout, nstvout, nstfout) every 100 steps (0.1 with your 1 fs 
timestep). It may be that writing this amount of data to disk is the 
limiting step in your case, and in any case it seems you are wasting 
disk-space (although that may not be a problem for you...).

In general, full precision x and v are required only for making 
re-starts, for nearly all analysis purposes, the reduced precision 
coordinates (nstxtcout) will be good enough!

Most people use a 2 fs timestep and constraints (like David suggested), 
and set nstxout and nstvout to something like 10000; nstfout you can set 
to zero.


-- 
Groetjes,

Anton

* NOTE: New Phone & Fax numbers (below) *

  _____________ _______________________________________________________
|             |                                                       |
|  _   _  ___,| K. Anton Feenstra                                     |
| / \ / \'| | | Dept. of Pharmacochem. - Vrije Universiteit Amsterdam |
|(   |   )| | | De Boelelaan 1083 - 1081 HV Amsterdam - Netherlands   |
| \_/ \_/ | | | Tel: +31 20 59 87608 - Fax: +31 20 59 87610           |
|             | Feenstra at few.vu.nl - www.few.vu.nl/~feenstra/         |
|             | "If You See Me Getting High, Knock Me Down" (RHCP)    |
|_____________|_______________________________________________________|