On Fri, Feb 12, 2021, Domenico Marson wrote:
>
>As far as I understand this is possible in GROMACS, so I think it should be
>possible also in AMBER, I just don't understand how to achieve the wanted
>behaviour (serial GPU AMBER run + MPI PLUMED run).
The only straightforward path is to use the openMP capabilities of plumed,
so MPI doesn't get involved. Basic info is here:
https://www.plumed.org/doc-v2.5/user-doc/html/_openmp.html
Warning: I haven't tried this, but it seems like it ought to work. But only
parts of plumed use openMP, so it may not help for your particular use case.
Making a version of pmemd.cuda.MPI where only thread 0 interacts with the
GPU sounds possible, but also sounds like a fair amount of coding work.
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Feb 18 2021 - 05:30:05 PST
