Nwchem

NWChem

The NWChem version 6.0 package was built using the Intel v12.1 compilers, and the IntelMPI v.4 library, on CentOS 6.

The following environment was used:

<source lang="bash">

1. environment variables needed to build NWChem for the GPC

1. must make sure to load modules gcc, intel, intelmpi

module load gcc intel intelmpi

export LARGE_FILES=TRUE export NWCHEM_TOP=/scinet/gpc/src/nwchem-6.0 export TCGRSH=/usr/bin/ssh export NWCHEM_TARGET=LINUX64

1. use the BLAS in MKL

export HAS_BLAS=yes export BLASOPT="-L$MKLROOT/lib/intel64 -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core -lpthread"

1. export LIB_DEFINES='-DDFLT_TOT_MEM=16777216'

export USE_MPI=y export USE_MPIF=y export IB_HOME=/usr export IB_INCLUDE=$IB_HOME/include export IB_LIB=$IB_HOME/lib64 export IB_LIB_NAME="-libumad -libverbs -lpthread" export ARMCI_NETWORK=OPENIB export MPI_LOC=$MPI_HOME export MPI_LIB=$MPI_LOC/lib64 export MPI_INCLUDE=$MPI_LOC/include64 export LIBMPI='-lmpigf -lmpigi -lmpi_ilp64 -lmpi'

1. from the Argonne wiki https://wiki.alcf.anl.gov/index.php/NWChem

export NWCHEM_MODULES=all export FOPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops -unroll-aggressive" export COPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops"

</source>

The above script is /scinet/gpc/Applications/NWChem-6.0/utils/env.sh .

Important note:

NWChem must be run in "direct" mode. That is, recalculating integrals on demand, and NOT saving them to disk. This is especially important when running on a large cluster with a shared filesystem. In addition, it has been observed that direct mode is at least 2-3 times faster than using temporary files on disk. For example, when running an MP2 computation, the following option MUST be set:

  TASK DIRECT_MP2 optimize

Running NWChem

- Load the necessary modules: gcc, intel, intelmpi, nwchem (best done in your .bashrc)

  # module load gcc intel intelmpi nwchem

- Make sure you've made a link in your home directory to the default.nwchemrc file in the installation:

  # ln -s $SCINET_NWCHEM_HOME/data/default.nwchemrc ~/.nwchemrc

If you had used a previous version of nwchem, please remove the link and recreate it, or remove it and copy the default .nwchemrc to your home directory.

- The NWChem executable is in $SCINET_NWCHEM_HOME/bin/nwchem

- For multinode runs, use the sample IB script futher below

- Create a torque script to run NWChem. Here is an example for a calculation on a single 8-core node:

<source lang="bash">

1. !/bin/bash
2. PBS -l nodes=1:ppn=8,walltime=48:00:00
3. PBS -N nwchemjob

1. 1. To submit type: qsub nwc.sh (where nwc.sh is the name of this script)

1. If not an interactive job (i.e. -I), then cd into the directory where
2. I typed qsub.

if [ "$PBS_ENVIRONMENT" != "PBS_INTERACTIVE" ]; then

  if [ -n "$PBS_O_WORKDIR" ]; then
    cd $PBS_O_WORKDIR
  fi

fi

1. the input file is typically named something like "nwchemjob.nw"
2. so the calculation will be run like "mpirun nwchem nwchemjob.nw > nwchemjob.out"

1. load the nwchem and other required modules if not in .bashrc already

module load gcc intel intelmpi nwchem

1. run the program

mpirun -r ssh -env I_MPI_DEVICE ssm -np 8 $SCINET_NWCHEM_HOME/bin/nwchem nwchemjob.nw >& nwchemjob.out </source>

Here is a similar script, but this one uses 2 InfiniBand-connected nodes:

<source lang="bash">

1. !/bin/bash
2. PBS -l nodes=2:ib:ppn=8,walltime=48:00:00
3. PBS -N nwchemjob

1. 1. To submit type: qsub nwc.sh (where nwc.sh is the name of this script)

1. If not an interactive job (i.e. -I), then cd into the directory where
2. I typed qsub.

if [ "$PBS_ENVIRONMENT" != "PBS_INTERACTIVE" ]; then

  if [ -n "$PBS_O_WORKDIR" ]; then
    cd $PBS_O_WORKDIR
  fi

fi

1. the input file is typically named something like "nwchemjob.nw"
2. so the calculation will be run like "mpirun nwchem nwchemjob.nw > nwchemjob.out"

1. load the nwchem and other required modules if not in .bashrc already

module load gcc intel intelmpi nwchem

1. run the program

mpirun -r ssh -env I_MPI_DEVICE rdssm -np 16 $SCINET_NWCHEM_HOME/bin/nwchem nwchemjob.nw >& nwchemjob.out </source>

notes on the older NWChem, 5.1.1, built on CentOS 5

The NWChem version 5.1.1 package was built using the Intel v11.1 compilers, and the IntelMPI v.4 library.

The following environment was used:

<source lang="bash">

1. environment variables needed to build NWChem for the GPC

1. must make sure to load modules gcc, intel, intelmpi

module load gcc intel intelmpi

export LARGE_FILES=TRUE export NWCHEM_TOP=/scinet/gpc/src/nwchem-5.1.1 export TCGRSH=/usr/bin/ssh export NWCHEM_TARGET=LINUX64

1. use the BLAS in MKL - Actually, this was NOT used, as there is a problem with 8-byte integers
2. export HAS_BLAS=yes
3. export BLASOPT="-L$MKLPATH $MKLPATH/libmkl_solver_ilp64_sequential.a -Wl,--start-group -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core -Wl,--end-group -lpthread"
4. export LIB_DEFINES='-DDFLT_TOT_MEM=16777216'

export USE_MPI=y export USE_MPIF=y export IB_HOME=/usr export IB_INCLUDE=$IB_HOME/include export IB_LIB=$IB_HOME/lib64 export IB_LIB_NAME="-libumad -libverbs -lpthread" export ARMCI_NETWORK=OPENIB export MPI_LOC=$MPI_HOME export MPI_LIB=$MPI_LOC/lib64 export MPI_INCLUDE=$MPI_LOC/include64 export LIBMPI='-lmpigf -lmpigi -lmpi_ilp64 -lmpi'

1. from the Argonne wiki https://wiki.alcf.anl.gov/index.php/NWChem

export NWCHEM_MODULES=all

1. export CCSDTQ=y
2. export CCSDTLR=y

export FOPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops -unroll-aggressive"

1. export FOPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops -multiple-processes=8 -unroll-aggressive"

export COPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops" </source>

The above script is /scinet/gpc/Applications/NWChem-5.1.1/utils/env.sh .

-- dgruner 3 September 2010