From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Fri Jun 21 2013 - 03:11:07 CDT
Hi mihaela,
+idlepoll is _not_, as posted before, related to CUDA (but is forced for it)
and is available when namd has been compiled against charm++ instead of a
separate MPI, as you seem to have done it. OpenMPI for example has the
idlepoll behavior by default, other MPIs maybe have options for it. It's
sometimes a gain of 2x when using +idlepoll also in CPU environments.
Idlepoll reduces the communication latency while polling for new messages
that often, that the CPU won't get idle and the messages are noticed as soon
as possible.
Norman Geist.
Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag
von Aron Broom
Gesendet: Donnerstag, 20. Juni 2013 17:20
An: Mihaela Drenscko
Cc: namd-l_at_ks.uiuc.edu
Betreff: Re: namd-l:
you need to be using the CUDA version of the code to use that, so if you are
not, remove it. Or if you really have GPUs available, get the CUDA
versions.
On Thu, Jun 20, 2013 at 11:14 AM, Mihaela Drenscko <quo.physics_at_gmail.com>
wrote:
Hello,
I am running a simulation and I get the following error:
FATAL ERROR: Unknown command-line option +idlepoll
How do I fix it?
Thank you,
Mihaela
-- Aron Broom M.Sc PhD Student Department of Chemistry University of Waterloo
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:20 CST