On Tue, Sep 25, 2012 at 12:48 AM, Mary Varughese <maryvj1985.gmail.com>wrote:
> Hi,
>
> can PMEMD.cuda and sander.MPI be configured together on a GPU?
>
No. There is a separate configure and build stage for each of the
following installs:
1) Amber in serial (includes pmemd, sander, and all of AmberTools in serial)
2) Amber in parallel (includes pmemd.MPI, sander.MPI, and some AmberTools
in parallel, like MMPBSA.py.MPI)
3) CUDA in serial (JUST builds pmemd.cuda)
4) CUDA in parallel (JUST builds pmemd.cuda.MPI)
You can install any of them on any machine, provided that you meet the
prerequisites. In fact, you can even install the CUDA executables on a
machine with no CUDA-capable card as long as you have the Nvidia CUDA
compiler installed, you just can't run it anywhere. Only CUDA code can be
compiled to run on a GPU (as far as Amber is concerned). Therefore, if you
have a machine with 8 CPUs and 2 GPUs, you can run sander.MPI on any (to
all) of the CPUs and pmemd.cuda(.MPI) on either/both of the GPUs.
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 25 2012 - 05:00:03 PDT
