On Thu, Dec 12, 2019, Seibold, Steve Allan wrote:
>
>I started a run (GaMD pmemd.cuda) on a GPU system which ran fine for
>2ns. I then followed this using "all" the same input files on a CPU
>system (GaMD pmemd.MPI) and again, it ran fine for 2ns. However, I
>switched to another cluster system and attempted to run on their
>pmemd.MPI and I got "vlimit exceeded....Coordinate resetting cannot be
>accomplished, deviation is too large" within in the first step or so. So,
>the protein setup is stable on GPU and a separate CPU system, but not on
>a different Cluster CPU. Can someone inform me as to the problem?
I'd double/triple check that operator error (e.g. using the wrong input
files) is not somehow present. When you get a vlimit or shake error on
the first step, you should generally do this: run a short simulation
(say 10 steps) with ntpr=1. Look at the energies at the first step and
compare them to those on the final step of the previous run. Something
is changing upon the restart, and you need to track down what that is.
Of course, turn GaMD on and off with short runs to see if the problem
depends upon those settings. Try regular aMD vs GaMD. Debugging
computer problems is generally in exercise in winnowing down the number
of possible sources of error.
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Dec 13 2019 - 04:30:02 PST
