On Fri, Oct 5, 2012 at 7:48 AM, Carlos Simmerling <
carlos.simmerling.gmail.com> wrote:
> pmemd can be used on more cores than sander for GB calculations. not just
> better scaling, but sander has hard limits of # cores for GB (1 per
> residue, I think).
>
True, sander has that limit since the bond/angle/dihedral workload is
broken up by residue, but pmemd has a similar, but slightly smaller, limit
in place (10+ atoms per core or pmemd will complain). The main GB loop is
still parallelized exactly the same way in both programs, and so they even
scale similarly.
In fact, I think the *fastest* GB routines in Amber are in NAB*, since it
has a nonbonded pairlist for GB (although I can't speak on the parallel
scaling).
All the best,
Jason
*Fastest CPU routines
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Oct 05 2012 - 07:30:02 PDT
