[gmx-developers] Posit-enablement of GROMACS

[Mark Abraham]

Hi,

On Tue, Jul 31, 2018 at 6:14 AM Theodore Omtzigt <theo at stillwater-sc.com>
wrote:

> Dear Devs:
>
>
> I am trying to connect with the current developers/maintainers of the
> GROMACS code. We have been working with a large international consortium of
> computational scientists on a new number system, called posits, to replace
> IEEE floating point with a high-performance reproducible tapered floating
> point format with higher numerical accuracy.
>

That sounds interesting, but I suspect there is not currently a compelling
advantage for molecular dynamics applications. Fidelity in computation is
not known to be a dominant source of error in the simulations, whereas
doubts about the accuracy of the model physics ("force field") and the
limitations created by finite sampling are well known. And the latter
implies a strong need to use the fastest form of computation that can
implement models that can be shown to be consistent with real physics. If a
change of representation facilitated the evaluation of a 1/sqrt(x) function
that on commodity hardware had an execution time similar to multiplication,
then that would be interesting. But without hardware support, the impact of
such an implementation for GROMACS users would be zero.

One of our supercomputer center customers want to enable GROMACS with this
> new number system, and I am looking for subject matter experts that we
> should work with to enable GROMACS with the C++ posit arithmetic library
> that has been developed for the posit-based hardware accelerators.
>

What advantage has that center identified?

There are "plain C" versions of the various numerical kernels that are
already available to be the basis for such efforts. So it would "simply" be
a matter of converting from IEEE to posit before entering the main loop and
calling those kernels within the loop. Early implementations should only
consider doing the force computation in posit representation.

Mark

Looking forward to connecting with the computational kernel SMEs of GROMACS.
>
> With warm regards,
>
> Theodore Omtzigt
>
> --
> 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/20180731/f0332ee7/attachment.html>