From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Wed Jan 18 2017 - 01:10:26 CST
By the way, NAMD is kind enough to print the ns/day right after the initial
load balancing, so one can quickly see the impact of any configuration
changes. In contrast this step is horrible with gromacs, since it does not
output any timing information during the run.
> -----Ursprüngliche Nachricht-----
> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
> Auftrag von Vermaas, Joshua
> Gesendet: Dienstag, 17. Januar 2017 17:48
> An: namd-l_at_ks.uiuc.edu; Nikhil Maroli <scinikhil_at_gmail.com>
> Betreff: Re: namd-l: NAMD performance monitor
>
> Hi Nikhil,
>
> NAMD doesn't have the same sort of accounting that Gromacs does. Instead,
> I tend to just try a bunch of different combinations, and use grep to get
> "Benchmark" lines from the log file, which has days/ns as one of its
outputs.
>
> -Josh
>
> On 01/17/2017 05:49 AM, Nikhil Maroli wrote:
> Dear all,
>
> We are having a workstation with 3 x GTX 1070, 2 x 12 core processor and i
> have used the following command to execute the job
>
> ./namd2 +p46 +setcpuaffinity +devices 0,2,3 new.inp > log.log &
>
> I was wondering is there any way to understand the performance details
> such as ns/day or the best combination of cpu-gpu. When i was working with
> Gromacs, at the end of the log file i could get the performance details
based
> on that we can try different combinations of cpu-gpu to get better
> performance.
>
> It will be helpfull if some one can suggest me how to find the best
> combination for better performance or the command to use for running job
> other than above mentioned.
>
> --
> Regards,
> Nikhil Maroli
>
>
This archive was generated by hypermail 2.1.6 : Sun Dec 31 2017 - 23:21:00 CST