[gmx-users] At what point is the random seed generated?

Szilárd Páll pall.szilard at gmail.com
Wed Jan 29 13:37:24 CET 2014


On Wed, Jan 29, 2014 at 4:46 AM, Trayder Thomas
<trayder.thomas at monash.edu> wrote:
> Curses! Kinda seemed intuitive that way. I assume the philosophy being
> worked towards is that every unique run should have a unique .tpr file that
> it is reproducible from?
>
> Curiously, I ran and checked a first batch on a different cluster and the
> simulations rapidly diverged (even between simulations run one after the
> other on the same node), while the second cluster managed to create 50
> identical simulations across 50 different nodes. The most obvious
> difference being that the diverging simulations used GPUs.

GPU runs will naturally diverge as we use atomic operations for the
final step in force reduction/accumulation, hence the implementation
is inherently non-deterministic. Your runs on the "second cluster"
must have used no-domain decomposition, though, otherwise those runs
would also be non-deterministic.

>
> I suspect I never would have noticed if I had of stuck to the one cluster.
> Might be worth stating explicitly in the manual for gen-seed, with
> reproducibility being what it is.
>
> Oh well, Thanks guys.
>
> -Trayder
>
>
>
>
> On Tue, Jan 28, 2014 at 10:55 PM, Mark Abraham <mark.j.abraham at gmail.com>wrote:
>
>> Indeed. grompp does report the seed it uses to stderr.
>>
>> Mark
>>
>>
>> On Tue, Jan 28, 2014 at 8:51 AM, David van der Spoel
>> <spoel at xray.bmc.uu.se>wrote:
>>
>> > On 2014-01-28 08:37, Trayder Thomas wrote:
>> >
>> >> Hi,
>> >> When using 'gen-seed = -1' at what point is the random seed assigned?
>> >>
>> >> e.g. Does it use the process ID of grompp and embed the seed number in
>> the
>> >> tpr file, or does it use the process ID of mdrun?
>> >>
>> >> I ask because I have 50 identical simulations started from the same tpr
>> >> file :(
>> >>
>> >> Thanks,
>> >> -Trayder
>> >>
>> >>  I guess you found out. Sorry about that. It is done in grompp and same
>> > tpr should give reproducible results. So you want to generate 50
>> different
>> > tpr files.
>> >
>> > --
>> > David van der Spoel, Ph.D., Professor of Biology
>> > Dept. of Cell & Molec. Biol., Uppsala University.
>> > Box 596, 75124 Uppsala, Sweden. Phone:  +46184714205.
>> > spoel at xray.bmc.uu.se    http://folding.bmc.uu.se
>> >
>> > --
>> > 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.
>> >
>> --
>> 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.
>>
> --
> 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 mailing list