On Mon, Dec 10, 2018, Ibrahim M. Moustafa wrote:
> I'm trying to do MD simulations using AMBER18 for large systems in explicit
>solvent. Amber seems able to be able to handle a size of 1.09 million atoms
>very well.However, trying to minimize a system with over 2 million atoms did
>not work, giving the error message given below. The minimization tried using
>sander.MPI with -np 32. The computer running the simulations has 128 GB RAM and
>4x 1080Ti GPU.
First, use pmemd.MPI on a cpu. It is *much* better suited to large
simulations than in sander.MPI. You can also try pmemd.cuda, which
should be much faster if you have enough memory. (With modern cards
like the 1080Ti, don't try pmemd.cuda.MPI unless you have time on your
hands, and are prepared for poor scaling across multiple GPUs.)
(If you really need sander, editing MAXPR as suggested in the error
messages stands a good chance of fixing the problem.)
I've simulated systems with up to 3.5 million atoms with pmemd. Not
sure what the actual limit is, and it probably depends on hardware.
...good luck...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Dec 10 2018 - 11:00:03 PST
