On Tue, Nov 12, 2013 at 3:13 AM, X Sun <egeh00.aol.com> wrote:
> Dear all,
>
> I used parmED.py to modify some radius parameters, which means the A COEF
> are modified. The prmtop file worked well with sander.MPI. But when I tried
> to move my simulation to GPU to fasten it, I found some error prompt showed
> pmemd.cuda could not deal with prmtop file in which A COEF were modified.
> Is there any method to solve this problem? GPU is really fast, I hope I may
> run my simulaiton on it.
>
Did you use "changeLJPair" in ParmEd? This command, which is functionally
equivalent to "NBFIX" from the CHARMM world, changes both the ACOEF and
BCOEF for that particular pair of atom types.
That said, the GPU code does not (yet) support changing the off-diagonal
elements of the Lennard Jones matrix. Due to the small amount of available
shared memory, the original LJ radius and well depths are back-calculated
from the diagonal terms of the Lennard Jones coefficients and recombined
each step according to the Lorentz-Berthelot combining rules. As a result,
any topology file modified by "changeLJPair" or "changeLJ14Pair" in ParmEd
will not run with pmemd.cuda.
This may change in Amber 14, but I'm not sure. It definitely will not work
in Amber 12, though. You can use pmemd.MPI for improved performance over
sander.MPI, but for GPUs it will not work.
Hope this helps,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Nov 12 2013 - 17:30:02 PST
