Re: [AMBER] pH-REMD GPU runs

From: Adrian Roitberg <>
Date: Wed, 2 Dec 2015 09:01:16 -0500

Dear Arsh

Yes, each replica will need its own GPU, sorry. Putting more than one
replica in a single GPU might be possible for small systems and a LOT of
work, but I honestly do not think we will do it.

While technically the constant pH (pH-remd) code can run both in cpus
and gpus, in reality you do not get a huge boost from moving to the GPU.
This is simply because the code as written in Amber 12 had the GB
sections done in the CPU, which was not fast and slowed things down.

Amber 16, scheduled to be released in April next year, will have a full
implementation of constant pH and pH-remd in GPU. We have been testing
and benchmarking in house and we are very happy with the results.


On 12/2/15 8:53 AM, Arjun Sharma wrote:
> Hi,
> Iím struggling to get GPU-accelerated explicit solvent pH-replica exchange jobs to run. I have access to 2 NVIDIA Tesla K20x GPUís.
> Pmemd CUDA-enabled mpi build runs just fine for non-REMD runs and provides great acceleration to my simulations. However, this doesnít work for my 12-replica pH-REMD runs. Does each replica needs its own GPU ? or can GPU be shared by multiple replicas ? Is there a way I can divide 2 GPU nodes available to run 12 replicas ?
> Thanks,
> Arsh
> _______________________________________________
> AMBER mailing list

Dr. Adrian E. Roitberg
Department of Chemistry
University of Florida
AMBER mailing list
Received on Wed Dec 02 2015 - 06:30:02 PST
Custom Search