[gmx-users] Can I expect numerical reproducibility between Gromacs 4.6 and Gromacs 5
mark.j.abraham at gmail.com
Fri Aug 1 18:14:32 CEST 2014
The RNG implementation is different, so for that aspect the answer is
"definitely not." Otherwise, you can probably find many pockets of
algorithm combination space that are sufficiently unchanged over 4.6->5.0
that you can observe such numerical reproducibility, but it was certainly
not a design feature. The SIMD implementation was revamped with an eye to
future portability, and the order of operations there could well have
changed, so try turning that off if you want/need to.
On Fri, Aug 1, 2014 at 10:37 AM, ms <devicerandom at gmail.com> wrote:
> Dear all,
> Me and a collegue were wondering if -using the same identical set of
> features, same input files, same random seed, running on a single CPU- we
> could expect numerical reproducibility between Gromacs 4.6 and Gromacs 5.0
> ; that is, same output numbers and same trajectory. My bet is on "no", but
> we would like to be sure. If there is a subset of features where this
> happens, this would be interesting to know.
> Gromacs Users mailing list
> * Please search the archive at http://www.gromacs.org/
> Support/Mailing_Lists/GMX-Users_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-users or
> send a mail to gmx-users-request at gromacs.org.
More information about the gromacs.org_gmx-users