On Wed, Mar 17, 2010 at 3:55 AM, Mark Abraham <mark.abra...@anu.edu.au>wrote:
> On 17/03/2010 6:19 PM, BIN ZHANG wrote: > >> Dear Mark: >> >> Thanks for your reply and sorry for my less info. >> > > I'm confused about your statement "I was trying to build gromacs 4.0 on a > cray-xt4 machine, the same one as the benchmark in the gromacs4 paper" > because I can't see any mention of results on such a machine in J. Chem. > Theory Comput. 2008, 4, 435-447 > > > Here is the configuration line I used: >> module load fftw/2.1.5.1 >> module load libxml2 >> CXX=CC CC=cc F77=ftn MPICC=cc LIBS="$XML2_LIBDIR -L/usr/X11/lib64" >> ./configure --enable-mpi --enable-double --enable-fortran >> --prefix=$HOME/gmx/para --disable-nice --with-fft=fftw2 >> --program-suffix=_mpi >> > > The use of double precision and the absence of hardware-optimized inner > loops (--enable-fortran, since AFAIK there's no alternative on XT4) will > slow things down compared to an Intel machine. No. These are standard AMD CPUs and thus SSE2 works find. I use module swap PrgEnv-pgi PrgEnv-gnu module swap gcc gcc/4.2.4 ./configure '--enable-mpi' '--without-x' 'CC=cc' 'CPP=cpp' 'CXXCPP=cpp' 'CXX=CC' '--without-xml' 'CPPFLAGS=-I/sw/xt5/fftw/3.1.2/sles10.1_gnu4.2.4/include' 'LDFLAGS=-L/sw/xt5/fftw/3.1.2/sles10.1_gnu4.2.4/lib -lfftw3f' 'CFLAGS=-O3 -fomit-frame-pointer -finline-functions -Wall -Wno-unused -msse2 -funroll-all-loops -std=gnu99' > > > Also, I searched the maillist and added ddorder in the mdrun line, but >> that doesn't seem to help much. >> aprun -n 128 $gmxdir/mdrun_mpi -deffnm npt_02 -g npt_02.log -ddorder >> cartesian -npme 32 >> > > You will need to experiment with npme and ddorder to find what's best. Also > consider g_tune_pme in the git codebase. > If you want to scale to larger number of CPUs you probably need 2D PME. It is available in the 2dpme git branch. (That it is in a branch means it is at the moment considered experimental - even more than mast that is) Roland > > Mark > > > The system is a membrane protein and I'm using berger lipid/OPLS-AA >> force filed. >> >> Thanks >> Bin >> >> >> On Mar 17, 2010, at 12:02 AM, Mark Abraham wrote: >> >> On 17/03/2010 5:43 PM, BIN ZHANG wrote: >>> >>>> Dear all: >>>> >>>> I was trying to build gromacs 4.0 on a cray-xt4 machine, the same one as >>>> the benchmark in the gromacs4 paper (). However, my timing for a similar >>>> system, ~100,000 atoms, is almost 3 times slower than in the paper. For >>>> example, I only got ~25 ns/day with 128 cpus. So is there any special >>>> flags, or tricks I can use during the configuration for detailed tuning? >>>> >>> >>> Probably. You've made it hard to help you when you haven't provided >>> (at least) your configure command, compiler and version, mdrun command >>> line and simulation system composition. >>> >>> Mark >>> -- >>> gmx-users mailing list gmx-users@gromacs.org >>> http://lists.gromacs.org/mailman/listinfo/gmx-users >>> Please search the archive at http://www.gromacs.org/search before >>> posting! >>> Please don't post (un)subscribe requests to the list. Use the www >>> interface or send it to gmx-users-requ...@gromacs.org. >>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php >>> >> >> -- > gmx-users mailing list gmx-users@gromacs.org > http://lists.gromacs.org/mailman/listinfo/gmx-users > Please search the archive at http://www.gromacs.org/search before posting! > Please don't post (un)subscribe requests to the list. Use the www interface > or send it to gmx-users-requ...@gromacs.org. > Can't post? Read http://www.gromacs.org/mailing_lists/users.php > -- ORNL/UT Center for Molecular Biophysics cmb.ornl.gov 865-241-1537, ORNL PO BOX 2008 MS6309
-- gmx-users mailing list gmx-users@gromacs.org http://lists.gromacs.org/mailman/listinfo/gmx-users Please search the archive at http://www.gromacs.org/search before posting! Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-requ...@gromacs.org. Can't post? Read http://www.gromacs.org/mailing_lists/users.php
