Re: [AMBER] Compare the result between pmemd.cuda and pmemd.mpi?

From: Jason Swails <>
Date: Mon, 02 Dec 2013 13:49:42 -0500

On Mon, 2013-12-02 at 12:11 -0600, X Sun wrote:
> Dear all,
> I am using pmemd to run MD simulation now both on GPU and CPU, depending on which are avaible on the HPC. My advisor asked me to compare the result of GPU and CPU to make sure the result of GPU is relable. The MD simulation is random, so comparing the result of MD simulation is pointless, is there any method for the comparision?

Simulations run on the GPU will not be identical to those run on the CPU
for reasons you seem to already understand. However, ensemble
properties (of converged ensembles) should be identical, and
finite-sized ensembles should yield the same results within the
uncertainty of the measurement for both ensembles. Pick a bunch of
properties (dihedral angle distributions, radial distribution functions,
radius of gyration, etc.) and make sure that the distributions (i.e.,
average properties) match for the CPU and GPU simulations. (You have to
make sure to run long enough with both to get converged ensembles,
though). The more properties you match, the better.

This is discussed in the relevant GPU citations found on the Amber
website and in the output of pmemd.cuda, I believe.


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Mon Dec 02 2013 - 11:00:03 PST
Custom Search