Dear Prof. Case,
thank you very much for your suggestions.
Using a slightly bigger box worked well for me.
Best regards,
David Schroeder
-----Ursprüngliche Nachricht-----
Von: David A Case <david.case.rutgers.edu>
Gesendet: Dienstag, 22. Oktober 2019 14:17
An: AMBER Mailing List <amber.ambermd.org>
Betreff: Re: [AMBER] error while running pmemd.cuda after reinstallation of amber 16
On Tue, Oct 22, 2019, Schröder, David Christopher wrote:
>
>The system is a small 5 residue cyclic peptide in DMSO solvent box.
>The trunc octah. box has a size of 35.8 and contains 3043 Atoms.
>
>gpu_neighbor_list_setup :: Small box detected, with <= 2 cells in one or more
> dimensions.
>Can I safely proceed with the -AllowSmallBox flag or does this need to
>be patched?
My experience with -AllowSmallBox has been that problems still seem to occur, so I don't think it is "safe". What cutoff and skinnb values are you using? Setting them to 8.0 and 1.0 or 2.0 may allow the simulation to continue (I don't know exactly what the calculation is for a truncated octahedron.)
If that doesn't work, your two options are to make a bigger solvent box (i.e. more DMSO molecules) or use the CPU version of the code. You probably don't need a much bigger box, but you may have to experiment a bit to find out what will work.
Amber's GPU code optimizations are agressive, and have never allowed for small periodic unit cells. We really only have one developer (Dave
Cerutti) who groks this part of the CPU code, and the coding effort required to accommodate small box sizes has been deemed too great, given the relatively small demand for small systems and the potential workarounds listed above.
...regards...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Oct 23 2019 - 09:30:02 PDT
