[gmx-developers] Re: g_tune_pme

[Roland Schulz]

Hi,

one more thing: rlistlong isn't updated. Thus rlistlong becomes shorter than
rlist.

Roland

On Fri, Jun 18, 2010 at 6:42 PM, Roland Schulz <roland at utk.edu> wrote:

> Hi Carsten,
>
> on Cray the number of processors need to be specified by "-n" not by "-np".
> I'm wandering what the best fix is:
>
> - Should we make an extra env-variable to specify the switch "-np"/"-n"?
> - or make the MPIRUN env-variable a template e.g. "mpirun -np %proc" (would
> mean bOmit option could be remove, because we could check whether the
> template includes the placeholder)
> - or make the MPIRUN env-variable include the -np/-n switch but not a a
> placeholder, e.g "mpirun -np" (would still require bOmit option)
> - or add an additional  "-n" switch to g_tune_pme (thus the user could call
> g_tune_pme either with -n or with -np and this would then be used to call
> mpirun)
>
> Independent of this I would suggest to also add a -downfac for a down
> scaling factor. Because PME got faster in 4.5 people might use a too large
> coulomb cut-off and not a too small one.
>
> Roland
>
>
>
> --
> ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
> 865-241-1537, ORNL PO BOX 2008 MS6309
>



-- 
ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
865-241-1537, ORNL PO BOX 2008 MS6309
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20100618/edae2b77/attachment.html>