Re: [AMBER] pmemd.cuda Energy Minimization Slow

From: Ross Walker <>
Date: Sat, 6 Jul 2013 23:17:52 -0700

Hi Parker,

The run is taking 10seconds of compute time which is about what i'd expect but for sone reason it takes 210 seconds of wallclock time. Are you sure nobody else is running on the same machine / same GPU? Are you using a remote (and slow?) filesystem?

The run should take about 10 seconds but for some reason it is only running at about 5% efficiency on your machine.

A few things to try:

1. Make sure you have the latest Nvidia driver.
2. Install the latest version of amber, should report v12.3 for GPU.
3. Reboot and kick everyone else off the machine.
4. Run on a local filesystem.
5. Download the benchmark suite from and see how your machine fairs with that.

All the best

On Jul 6, 2013, at 22:04, Parker de Waal <> wrote:

> Hi AMBER Users!
> I'm currently running a step 1000 energy minimization, using steepest
> descent on a system comprised of 54782 atoms on a single Tesla K20 card,
> and the entire process seems very slow compared to the same setup in
> GROMACS. .rst provided here which contains settings and timings etc:
> If anyone can provide insight I would be extremely grateful.
> Best,
> Parker
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Sat Jul 06 2013 - 23:30:02 PDT
Custom Search