Le 02/07/2025 à 05:44, David A Case via AMBER a écrit :
> I don't think more equilibration is likely to help here: it looks like
> there
> is something (bad) going on with A100 cards. We have seen some
> examples of
> this in the past. I'm hoping someone on the list with access to A100
> cards
> will have time to look at this.
Dear all,
I have access to A100 cards, and encountered this weird behaviour in the
past, before
following precisely the protocole by Callum Dickson:
https://github.com/callumjd/AMBER-Membrane_protein_tutorial
The tricky part between my former parameters and his (I don't use HMR
for production)
was during the second heating step where the timestep is 1fs instead of 2fs.
Double check carefully his parameters and yours, and you may probably
also find the culprit.
If you wish to test on another setup (the card may be broken, after all,
or NVIDIA cuda at fault?),
just send a link on the list with minimal paramters (rst, prmtop and
md.in) so we can give you output
from elsewhere. I regularly do CPU job first (or GPU) on my machine,
then continue on the supercomputer
and with the aforementionned protocol, it works perfectly.
HTH,
Stéphane
--
Maître de conférences Hors Classe, PEDR, HDR
US2B, Nantes Université, CNRS, UMR 6286, Team Structural Bioinformatics, F-44000 Nantes
http://www.us2b.univ-nantes.fr/ - http://www.steletch.org
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jul 02 2025 - 03:00:02 PDT
