Re: [AMBER] T-REMD Reproducibility - Cluster v.s Desktop

From: Jason Swails <jason.swails.gmail.com>
Date: Fri, 25 Sep 2015 13:32:58 -0400

On Fri, Sep 25, 2015 at 1:04 PM, Carlos Simmerling <
carlos.simmerling.gmail.com> wrote:

> I don't know of any way to get them to exactly match unless it's the same
> cpu and same compiler, etc .


​Just to add on a little here -- this statement is true for sander and
pmemd.cuda; but only those. The case with pmemd is a little worse -- you
can NEVER guarantee that two REMD simulations will give identical results
since the MPI load balancing is dynamic​ (in sander, it is static so
reproducibility is possible). But as Carlos mentioned, only for the same
compiler (and compiler options) and CPU or GPU.

HTH,
Jason

-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 25 2015 - 11:00:03 PDT
Custom Search