[gmx-developers] reproducability of md sims

Shirts, Michael R. (mrs5pt) mrs5pt at eservices.virginia.edu
Tue Dec 2 06:26:10 CET 2014


> So one can have fully deterministic forces in MD by simply paying a couple of flops per force component in each inner loop.

If there was a way to overload functions in such a way as to make this a compile time option, it might be pretty useful for validation . . .

~~~~~~~~~~~~
Michael Shirts
Associate Professor
Department of Chemical Engineering
University of Virginia
michael.shirts at virginia.edu
(434) 243-1821

From: Mark Abraham <mark.j.abraham at gmail.com<mailto:mark.j.abraham at gmail.com>>
Reply-To: "gmx-developers at gromacs.org<mailto:gmx-developers at gromacs.org>" <gmx-developers at gromacs.org<mailto:gmx-developers at gromacs.org>>
Date: Sunday, November 30, 2014 at 5:40 AM
To: Discussion list for GROMACS development <gmx-developers at gromacs.org<mailto:gmx-developers at gromacs.org>>
Subject: Re: [gmx-developers] reproducability of md sims



On Sun, Nov 30, 2014 at 12:03 PM, Carsten Kutzner <ckutzne at gwdg.de<mailto:ckutzne at gwdg.de>> wrote:
Hi,

On 30 Nov 2014, at 11:17, Berk Hess <hess at kth.se<mailto:hess at kth.se>> wrote:

> Hi,
>
> There is documentation on this somewhere.
> Even when running with MPI and/or OpenMP simulations should be reproducible when ran on an identical run
But reductions of several summands into a sum do not yield a bit-identical result across runs.
Therefore I think at least with MPI and/or OpenMP we will never get bit-identical
trajectories unless the order of summation is fixed somehow for such operations.

True, but not the whole story. You can leave the order free if you pre-round the operands such that the result is provably insensitive to the order of the operations. See http://htor.inf.ethz.ch/publications/img/arteaga-fuhrer-hoefler-reproducible-apps-ipdps14.pdf. So one can have fully deterministic forces in MD by simply paying a couple of flops per force component in each inner loop.

Mark

Best,
  Carsten


> setup. The only things that usually matter, and which -reprod should turn off, are dynamic load balancing and FFTW kernel choice by timing.
> You can mail me your tpr and I can have a look.
>
> Cheers,
>
> BerkOn Nov 30, 2014 9:05 AM, David van der Spoel <spoel at xray.bmc.uu.se<mailto:spoel at xray.bmc.uu.se>> wrote:
>>
>> Hi,
>>
>> I'm trying to test reproducibility of simulations of liquids. The
>> simulation systems are flexible (no constraints) and run with the
>> -reprod flag, starting from the same tpr, on a single core. However
>> after a few hundred fs the energy starts to diverge.
>>
>> Are there factors impacting the reproducibility that are not controlled
>> by the -reprod flag?
>>
>> --
>> David van der Spoel, Ph.D., Professor of Biology
>> Dept. of Cell & Molec. Biol., Uppsala University.
>> Box 596, 75124 Uppsala, Sweden. Phone: +46184714205<tel:%2B46184714205>.
>> spoel at xray.bmc.uu.se<mailto:spoel at xray.bmc.uu.se>    http://folding.bmc.uu.se
>> --
>> Gromacs Developers mailing list
>>
>> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or send a mail to gmx-developers-request at gromacs.org<mailto:gmx-developers-request at gromacs.org>.
> --
> Gromacs Developers mailing list
>
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or send a mail to gmx-developers-request at gromacs.org<mailto:gmx-developers-request at gromacs.org>.

--
Gromacs Developers mailing list

* Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before posting!

* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists

* For (un)subscribe requests visit
https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers or send a mail to gmx-developers-request at gromacs.org<mailto:gmx-developers-request at gromacs.org>.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20141202/426779bf/attachment.html>


More information about the gromacs.org_gmx-developers mailing list