Dear Prof. Case,
thank you for your kind reply.
> Just to be sure, are you up to date with code updates? In particular,
> have you applied update.7, which fixed a memory problem with COM
> restraints on GPUs?
I used Amber18.12 and Amber17.15 and was able to reproduce the described
problem in both versions.
> If so, see if using fewer atoms in the COM mask fixes anything.
My COM masks were well within, what is specified in the manual to work.
The mask :1 corresponds to 92 atoms, and .P31 to 94.
Still, I tried reducing the number of atoms. Reducing the number of
phosphor atoms to 32 (:PE.P31), did not solve the issue.
Reducing the number of atoms for the molecule in the membrane to 38
(:1.C=) initially seemed to solve the issue, but after ~1000 steps again
restraint energies overflowed and the system crashed.
Reducing the number of atoms for the molecule further to two
(:1.C21,C22), allows running the system on GPUs for several nanoseconds
without crashing or overflowing so far.
Still, I am troubled, because i do not see any rationale behind what is
working, and what is not.
I will send you my files and would appreciate you having a look into my
problem and probably shed some light onto what is going wrong here.
Best regards,
David
On 3/9/19 3:01 PM, David Case wrote:
> On Wed, Mar 06, 2019, David Bickel wrote:
>> I encountered a problem running simulations of a membrane system with
>> distance restraints on GPUs.
>> I used distance restraints (fxyz=0,0,1) between the
>> COM of the molecule (:1) and the phosphor atoms of the lipids (@P31), in
>> order to keep the molecule in the center of the membrane.
> It's certainly possible that some problem exists on GPUs with COM
> restraints, especially of the mask .P31 averages over lots of atoms.
>
> Just to be sure, are you up to date with code updates? In particular,
> have you applied update.7, which fixed a memory problem with COM
> restraints on GPUs?
>
> If so, see if using fewer atoms in the COM mask fixes anything. If that
> doesn't help, we'll probably need to get files from you to try to
> reproduce the problem.
>
> ...good luck....dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
David Bickel
PhD Student
Computational Pharmaceutical Chemistry and Molecular Bioinformatics
Heinrich-Heine-Universität Düsseldorf
Institut für Pharmazeutische und Medizinische Chemie
Universitätsstr. 1
D-40225 Düsseldorf
Germany
Office: 26.23.00.22
Phone: +49 211 8112528
URL: http://cpclab.uni-duesseldorf.de
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 11 2019 - 05:30:02 PDT
