From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Wed Apr 08 2009 - 10:16:36 CDT
Hi Shayantani,
try compiling using icc instead of gcc, and editing your .arch files to
use the same compilation flags as we do (see
http://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?NamdAtTexas).
merging your code with the namd 2.7 tree, if possible without too much
trouble, will also let you take advantage of sse loops which have made
things even faster recently.
Best,
Peter
shayan_at_msu.edu wrote:
> Thanks Axel for your reply.
> I have build using the 2.6 source, which, I beleive, is the same as
> the executable available on Ranger.
> I need to compile NAMD myself since I have a code customized according
> to my need.
> I have also compared the timing with the linux-amd64 binary available
> from NAMD home. The timing for 16cpus (1 node) are as follows:
> 16 CPU: 0.2533 s/step 1.4659 days/ns (executable from ~tg455591 on Ranger)
> 16 CPU: 0.9072 s/step 5.2504 days/ns (NAMD_2.6_Linux-amd64 binary from
> NAMD home)
> 16 CPU: 0.9850 s/step 5.7000 days/ns (my compiled version)
> Using the right options might help me to get a better executable
> atleast close to the one in Ranger.
>
> Thanks,
> Shayantani
>
> Quoting "Axel Kohlmeyer" <akohlmey_at_cmm.chem.upenn.edu>:
>
> > On Wed, 2009-04-08 at 08:33 -0400, shayan_at_msu.edu wrote:
> >> Hello NAMD users,
> >> I need to have a NAMD executable that can run on 16cpus of 1 node on
> >> Ranger without using the MPI environment. Can anyone tell me about the
> >> optimal build options for Ranger?
> >> I have tried to compile using the following options:
> >> build charm++ net-linux-amd64 -O -DCMK_OPTIMIZE=1
> >> config tcl fftw Linux-amd64-g++
> >> I have used the fftw-linux-amd64 and tcl-linux-amd64 libraries from
> >> the NAMD home.
> >> But, the compiled version is several times slower than the executable
> >> from the directory ~tg455591/Linux-amd64-MVAPICH-icc/namd2.
> >> Any suggesstion that will help me to compile and get close to the
> >> optimal timing would be greatly helpful.
> >
> > before getting in way over your head, did you try the precompiled
> > amd64 version from the namd homepage? you should also make sure that
> > you compare executables generated from the same sources, i.e. that you
> > don't mix the recent beta release with the 2.6 version.
> >
> > axel.
> >
> >>
> >> Thanking all,
> >> Shayantani
> >>
> >> Shayantani Mukherjee
> >> MSU.
> >>
> >
> > --
> > =======================================================================
> > Axel Kohlmeyer akohlmey_at_cmm.chem.upenn.edu http://www.cmm.upenn.edu
> > Center for Molecular Modeling -- University of Pennsylvania
> > Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
> > tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
> > =======================================================================
> > If you make something idiot-proof, the universe creates a better idiot.
> >
> >
> >
>
>
>
> Shayantani Mukherjee
> Postdoctoral Research Associate
> Dept. of Biochemistry and Molecular Biology
> Michigan State University
> E-mail: shayan_at_msu.edu
>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:34 CST