[AMBER-Developers] pmemd.cuda.MPI error

From: Bill Miller III <billymiller.ufl.edu>
Date: Fri, 3 Dec 2010 16:38:56 -0500

Dear Developers,

I have recently been running MD with a recently checked out version of Amber
11 using pmemd and decided to switch over to using pmemd.cuda.MPI. I am
using the NVIDIA Tesla S1070 GPUs on NICS Lincoln. I have had no problems
running this simulation with pmemd or pmemd.cuda. However, when I tried
pmemd.cuda.MPI I came across an error in the mdout:

| ERROR: max pairlist cutoff must be less than unit cell max sphere
radius!

I also get ***** and NaN for several of the energies in the last step or
two. The simulation usually runs for 15-17 steps before this error is
written and pmemd.cuda.MPI exits. I have visualized the trajectory output
file and everything seems ok until the last two steps. The second to last
step has two or three coordinates that are not in the box, and the following
(and last) step shows the system has blown up and atoms are scattered in a
much larger area than the original box.

I have attached my input files along with the mdout file from the run. I can
also supply the netcdf output trajectory upon request. I was wondering if
this is an error that others have seen or can reproduce as a possible bug in
pmemd.cuda.MPI.

Thanks.

-Bill

-- 
Bill Miller III
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-6715



_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers

Received on Fri Dec 03 2010 - 14:00:03 PST
Custom Search