AMBER Archive (2006)
Subject: RE: AMBER: sander problem
From: Ross Walker (ross_at_rosswalker.co.uk)
Date: Thu Feb 16 2006 - 10:51:24 CST
Sander v8.0 uses a binary tree to communicate the coordinates to all MPI
threads and this requires that the number of threads be a power of two.
The code also has a hard limit on 128 processors. This can be removed quite
easily but the code is not really efficient beyond this number of
processors. You should check the speedup you get on 128 processors before
trying to go further.
PMEMD uses a space decomposition approach and so does have this limitation
as the coordinates do not need to be copied to all threads. If you are
running a type of calculation that pmemd supports - in v8.0 that is a
periodic boundary PME calculation then you should use the PMEMD code. If
however your calculation is not supported by pmemd or you want to modify the
code in some way then use sander but be aware of the limitations and the
fact that it won't scale as well.
All the best
| Department of Molecular Biology TPC15 |
| The Scripps Research Institute |
| Tel: +1 858 784 8889 | EMail:- ross_at_rosswalker.co.uk |
| http://www.rosswalker.co.uk <http://www.rosswalker.co.uk/> | PGP Key
available on request |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of
Sent: Thursday, February 16, 2006 07:58
Subject: AMBER: sander problem
1) I have installed parallel version of Amber8 in a Linux cluster (HP
superdome) which runs on Enterprise Linux and Intel fortran 90 compiler.
Installation and testing was successful. When i run sander in parallel, i
get an error message that "no of processors should be as power of 2 and not
greater than 128".
Can anyone kindly suggest me what could be the problem. Is it the nature of
parallel sander. But i dnt face this problem with PMEMD.
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu