As Peter said, the simplest approach is just k-point parallel which does not need OpenMP. The rate limiting steps are lapack calls which the Intel mkl is good for.
Whether mpi is good for a small cluster such as you have in not so clear. Peter has benchmarked the mpi version and his conclusion is that it is not useful; I have benchmarked on my cluster and I have consistently found the mpi versions to be faster than the Intel threading. I suspect that it really depends upon many details specific to the systems, for instance cross-core optimizations, network and memory speed.... Not a complete answer, but for certain the simplest approach is to just use k-point parallel. On Fri, Feb 22, 2013 at 9:55 AM, Long Zhang <lzhang6789 at gmail.com> wrote: > > That's true I agree. If the k-point parallel version does not need mpi, then > it's OpenMP I suppose (please correct me if it's not). But I really don't > want to run OpenMP because it will mess my other applications up. > > I am building it on a small server of 2 Opteron 12-core cpus. It's just a > dual-cpu system, not a cluster. I am not sure if I can gain enough speed-up > by using 24 cores. Anyway if I want a little k-point parallel, can I choice > to use mpi? or there is no such choice at all? > > Thank you very much. > Long > > > On Fri, Feb 22, 2013 at 3:07 AM, Peter Blaha <pblaha at theochem.tuwien.ac.at> > wrote: >> >> I keep saying this all the time: >> >> The mpi-parallel version needs special hardware ! >> >> It is not of much use on a single quadcore-cpu. >> A cluster needs a fast network like infiniband. >> >> Only then the installation of the mpi-version make sense. >> >> We have a k-point parallel version, which does not need mpi, but is >> perfect for a small cluster with a common NFS filesystem. >> >> >> Am 22.02.2013 06:20, schrieb Long Zhang: >>> >>> >>> Hello, >>> >>> I am trying to install the current version 12.1 on Ubuntu 12.04 >>> (64-bit). I met some problem related to mpi, can someone please give me >>> some hint? Thanks in advance. >>> >>> The situation is like this, on a freshly installed Ubuntu 12.04 64bit, I >>> installed current version Intel Fortran Composer XE 2013.2.146, and >>> installed MPICH2 and FFTW3 via Ubuntu Software Center. >>> >>> Before run the siteconfig_lapw, I checked >>> which ifort >>> /opt/intel/composer_xe_2013.2.146/bin/intel64/ifort >>> which mpif90 >>> /usr/bin/mpif90 >>> >>> and did the following 3 lines in the terminal >>> source /opt/intel/composer_xe_2013.2.146/bin/compilervars.sh intel64 >>> export >>> >>> LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/intel/composer_xe_2013.2.146/mkl/lib/intel64 >>> export MKLROOT=/opt/intel/composer_xe_2013.2.146/mkl/ >>> >>> Then I run the siteconfig_lapw, and selected system as "ifort12+mkl" >>> (the first one), selected compilers as "ifort" and "cc", selected >>> "finegrained parallel" and "mpif90", and got error message like this: >>> >>> ----------------------------------------------- cut1 >>> ----------------------------------------------- >>> Compile time errors (if any) were: >>> SRC_hf/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_hf/compile.msg:make: *** [rp] Error 2 >>> SRC_hf/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_hf/compile.msg:make: *** [cp] Error 2 >>> SRC_lapw0/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_lapw0/compile.msg:make: *** [para] Error 2 >>> SRC_lapw1/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_lapw1/compile.msg:make: *** [rp] Error 2 >>> SRC_lapw1/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_lapw1/compile.msg:make: *** [cp] Error 2 >>> SRC_lapw2/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_lapw2/compile.msg:make: *** [rp] Error 2 >>> SRC_lapw2/compile.msg:make[1]: *** [modules.o] Error 1 >>> SRC_lapw2/compile.msg:make: *** [cp] Error 2 >>> ----------------------------------------------- cut1 >>> ----------------------------------------------- >>> >>> I looked into SRC_lapw1/compile.msg, and the detailed error is: >>> >>> ----------------------------------------------- cut2 >>> ----------------------------------------------- >>> mpif90 -FR -mp1 -w -prec_div -pc80 -pad -ip -DINTEL_VML -DFFTW3 >>> -traceback -DParallel -c modules_tmp_.F >>> >>> -------------------------------------------------------------------------- >>> The Open MPI wrapper compiler was unable to find the specified compiler >>> gfortran in your PATH. >>> >>> Note that this compiler was either specified at configure time or in >>> one of several possible environment variables. >>> >>> -------------------------------------------------------------------------- >>> make[1]: *** [modules.o] Error 1 >>> make[1]: Leaving directory `/home/lzhang200/wien2k/SRC_lapw1' >>> make: *** [rp] Error 2 >>> ----------------------------------------------- cut2 >>> ----------------------------------------------- >>> >>> It seems OpenMPI was invoked and it looked for gfortran. I don't have >>> gfortran installed at all, but I found I have openmpi installed. I then >>> used Ubuntu Software Center to search openmpi and uninstalled related >>> packages. I restarted the system and repeated the siteconfig_lapw. >>> However I still get error message, not same but similar, it still looks >>> for gfortran, for examplethe SRC_lapw1/compile.msg shows: >>> >>> >>> ----------------------------------------------- cut3 >>> ----------------------------------------------- >>> mpif90 -FR -mp1 -w -prec_div -pc80 -pad -ip -DINTEL_VML -DFFTW3 >>> -traceback -DParallel -c modules_tmp_.F >>> /usr/bin/mpif90: line 370: gfortran: command not found >>> make[1]: *** [modules.o] Error 127 >>> make[1]: Leaving directory `/home/lzhang200/wien2k/SRC_lapw1' >>> make: *** [rp] Error 2 >>> ----------------------------------------------- cut3 >>> ----------------------------------------------- >>> >>> I did not specify to use gfortran anywhere in the setupof >>> siteconfig_lapw, I don't see why mpif90 keeps looking for it. How can I >>> fix this problem? Thanks a lot. >>> >>> Regards, >>> Long Zhang >>> >>> >>> _______________________________________________ >>> Wien mailing list >>> Wien at zeus.theochem.tuwien.ac.at >>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien >>> >> >> -- >> Peter Blaha >> Inst.Materials Chemistry >> TU Vienna >> Getreidemarkt 9 >> A-1060 Vienna >> Austria >> +43-1-5880115671 >> >> _______________________________________________ >> Wien mailing list >> Wien at zeus.theochem.tuwien.ac.at >> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien > > -- Professor Laurence Marks Department of Materials Science and Engineering Northwestern University www.numis.northwestern.edu 1-847-491-3996 "Research is to see what everybody else has seen, and to think what nobody else has thought" Albert Szent-Gyorgi