From: Francesco Pietra (chiendarret_at_gmail.com)
Date: Wed Sep 11 2013 - 09:24:59 CDT
Hello:
Recently I posted here that a debug 16-replicas T-remd (polypeptide in
TIP3, 55118 atoms) crashed because of "stray PME charges". At different
step in different fresh attempts. It had been run with 32 tasks per node,
1024 total tasks.
No answer here, however I solved my problem by increasing to total 2048
tasks (I had 64nodes/16proc each).
As the system requires many replicas, I have now attempted a debug run with
64 replicas, in a doubled range of temperature, with same 32 tasks per
node, for 2048 total tasks. Crashed at step 580 for "Stray PME charges".
Does that again mean insufficient resources? Given the uncertainty, I did
not try with more resources because the above resources are at the limit
allowed for a debug run. More importantly, if it is a problem of resources,
it may easily become impossible to get to >200 replicas, unless one got a
fortune.
On the other hand, it was reported that AMBER, by setting igb=8, runs remd
under GB conditions faster and with better results than in explicit water.
Therefore, my question: is now remd namd2.9 under GB a routine procedure?
Fast enough to allow remd? It would shorten the number of replicas needed.
I was unable to catch a discussion on this possibility, however, in the
past, I was unable to run MD fast enough with namd under GB.
Thanks for clarify
francesco pietra
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:39 CST