On Mon, 2014-02-17 at 12:58 -0600, psu4.uic.edu wrote:
> Dear Amber developers,
>
> If not too much asking, wonder Amber 14 might be released this Spring
> (April), following the bi-annual update cycle? Also wonder what might be
> the exciting new functions like GPU supports on MM/PBSA and /or pmemd.MPI
MMPBSA.py will most likely never support GPUs. As Ross pointed out,
there is no currently available GPU-PB solver, so the only calculation
that is currently even possible to accelerate given existing
technologies would be GB -- the fastest option available already.
Furthermore, since CPU-GPU communications will have to be done for every
frame, the performance you see in pmemd.cuda is not directly
transferable to MMPBSA.py, even if it used GPUs to accelerate the GB
part. Part of what makes pmemd.cuda "the fastest MD engine in the world
on commodity hardware" is the fact that it very rarely has to
communicate data between the CPU and GPU.
The moral of the story, MM/PBSA will be CPU-bound until (or unless) a
GPU-accelerated PB solver is written and integrated with Amber.
All the best,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Feb 17 2014 - 13:00:03 PST
