On Tue, May 24, 2011, Jason Swails wrote:
> >
> > So, AMBER can utilise effectively 4 nodes with 16 cores per node for one
> > job? Am I right?
Let me amplify or simplify some of the earlier replies:
1. To get good parallel scaling, you must be running pmemd.MPI, *and*
the interconnect between nodes must be of high quality (e.g. infiniband
or equivalents). On a machine with just gigabit ethernet connections,
simulations are effectively limited to a single node.
2. pmemd.MPI indeed runs "well" on Teragrid machines, but with an upper limit
on the number of cores that can be used efficiently; this upper limit depends
on the particular machine and the size of the system; benchmark pages can help a
lot here.
.....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 25 2011 - 02:30:02 PDT
