Re: [AMBER] Differences in Lennard-Jones interactions in CPU vs GPU pmemd?

From: David A Case <david.case.rutgers.edu>
Date: Sun, 5 Apr 2020 13:15:15 -0400

On Mon, Mar 30, 2020, Hanne Antila wrote:
>
>
>wrapping first mol.: -2603377.53130 23007489.83260 3909449.84430

This looks very odd to me. Are there other "wrapping" messages in your
outputi file? Do any other also have gigantic numbers, like 2-3
million?

Have you visualized both the CPU and GPU trajectories? Does everything
look both OK and similar (up until the time the GPU job crashes)?

If so, it might be worth trying a GPU job with iwrap=0.

>
>The force field defines some special 1-4 interactions. Reading the
>mailing list, it seems to imply that these used to be treated differently
>in cpu vs gpu. Is this still the case? Or are there any other possible
>reasons why LJ interactions would be different in the two cases?

Do any of these special interactions involve atoms with zero LJ
parameters (which would be hydrogens)? It's possible that using the
"lmod" action in parmed could reduce the instabilities. But try the
other suggestions above first.

....dac


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Apr 05 2020 - 10:30:02 PDT
Custom Search