Re: [AMBER] Explicit solvent CpHMD - Are protonation state evaluations done on the GPU in Amber 16?

From: Jason Swails <jason.swails.gmail.com>
Date: Thu, 27 Oct 2016 15:58:59 -0400

On Thu, Oct 27, 2016 at 4:13 AM, Eric Lang <eric.lang.bristol.ac.uk> wrote:

> Hello,
>
> I would like to run some constant pH MD / pH replica exchange MD in
> explicit solvent.
>
> In Jason's tutorial on explicit solvent CpHMD (
> http://jswails.wikidot.com/explicit-solvent-constant-ph-md) which is based
> one Amber14 it is mentioned that "*c**onstant pH calculations are
> implemented on GPUs, but the GB protonation state evaluations are done on
> the CPU, and those are so slow that running ~8 or more processors is
> faster.*"
>

​As Adrian mentioned, this limitation is gone in Amber 16.​

​I updated my wiki to reflect that.​

I was wondering if it was still the case with Amber16 or if the GB
> protonation state evaluations are now done on the GPU as well and those
> CpHMD/pH-REMD are now faster to run on GPUs?
>

​The caveat that you shouldn't run multiple replicas on the same GPU
applies here as well. So if you have enough GPUs to assign one to each
replica, GPUs are definitely faster. If not, large CPU clusters may still
be better.

Thanks for the report,
Jason



>
> Many thanks in advance,
>
> Eric
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>



-- 
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Oct 27 2016 - 13:00:02 PDT
Custom Search