[gmx-developers] FMM in GMX-5.x

Mark Abraham mark.j.abraham at gmail.com
Thu Aug 14 22:42:51 CEST 2014


On Thu, Aug 14, 2014 at 1:42 PM, Smart Eagle <wangykoo at gmail.com> wrote:

> Hi,
>     I just notice that GMX have plan to implement FMM in to GMX to
> calculate long-term forces for MD. You know PME will suffer FFT NxN all
> communactiom problems and FMM seemly can scale much better than FME. Indeed
> there have been a progrom called MODYLAS which have implement FMM to MD (J.
> Chem. Theory Comput. 2013, 9, 3201−3209‍).
>     I just want to know therotically, which is larger between FMM and PME
> for the time cost of long-term forces calculation in one MD steps in one
> nodes ? (means the basic calculation load for each method without
> consideration of communications between nodes). ‍‍‍
>

Theoretically - it depends on the size of the simulation (which fixes N)
and the quality of the implementations (which fix the prefactors in the
asymptotic analysis of O(N) vs O(N log N)).

Practically - you could analyze the total flops on a given problem, but it
would tell you little about the performance on a single node, because how
well the implementation handles data flow is what dominates performance
(now, and more so in the future). Adding inter-node parallelism makes that
worse.

Mark


> cheers up
>
>
> --
> 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.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20140814/9d5f73b7/attachment.html>


More information about the gromacs.org_gmx-developers mailing list