From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Sun Feb 09 2014 - 17:13:15 CST
On Sun, Feb 9, 2014 at 4:05 PM, Mare Libero <mareliber27_at_yahoo.com> wrote:
> Hello,
>
> I noticed that the performances of MD simulations with namd is somewhat
> slower when IMD is turned on. I was wondering if this is normal, or if I did
> something wrong in setting up the parameters. I am running
> NAMD_2.9_Linux-x86_64-multicore-CUDA and my IMD configuration is:
>
> IMDon yes
> IMDport 3000 ;# port number (enter it in VMD)
> IMDfreq 1 ;# send every 1 frame
this statement may be causing the problem. updating every frame is
rarely needed and will force NAMD to spend a significant amount of
time collecting the coordinate info and sending it around. if you run
VMD on the same machine as NAMD, you also have to consider the amount
of resources consumed by VMD, especially when using the GPU (even if
it is not shared with NAMD). VMD will drop frames it cannot visualize
anyway, so you should experiment, which is a good setting for your
case.
> IMDwait yes ;# wait for VMD to connect before running?
>
> The difference in speeds is not large (may be less than 10%), but for long
> simulations it will make the difference.
i don't understand. if you care about MD performance, you should not
hook up IMD. or at least not in "wait" mode and/or not all the time.
otherwise, 10% don't really matter, since typically the problem is
time *scales*, i.e. that you would need to run the MD orders of
magnitude faster to see (and potentially manipulate) the process that
you are interested in with a reasonable speed.
axel
> Thanks in advance,
>
> Al
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 College of Science & Technology, Temple University, Philadelphia PA, USA International Centre for Theoretical Physics, Trieste. Italy.
This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:20:27 CST