NWChem is a quantum chemistry and molecular dynamics software which provides comprehensive many methods such as Hartree-Fock (HF), post-Hartree-Fock (post-HF), Density Functional Theory (DFT), classical molecular dynamics (MD), and hybrid quantum mechanics/molecular mechanics (QM/MM) for computing the properties of molecule. NWChem is an open-source, fast, efficient, and powerful. The program algorithm is optimized and designed to run with distributed-memory parallel method on Linux cluster across multiple compute nodes.
Website: http://www.nwchem-sw.org Github: https://github.com/nwchemgit/nwchem Documentation: https://github.com/nwchemgit/nwchem/wiki
- I compiled and installed NWChem 6.8.1 and 7.0.0 beta 1 on Red Hat 8, CentOS 7, Ubuntu 16.04, and Ubuntu 18.04 with and without enabling GPU (CUDA).
- Instance types
- CPU instance: t2.micro (free tier)
- CPU instance: c5.xlarge - 4 CPU cores
- CPU instance: c5.4xlarge - 16 CPU cores
- GPU instance: g3s.xlarge - 4 CPU cores + 1 GPU
- Storage
- SSD
- 30 GiB (free tier)
- C++ and Fortran compilers such as Intel or GNU compiler
- MPI library such as Intel MPI or OpenMPI
- Math library such as Intel MKL (BLAS, LAPACK, and ScaLAPACK) or GNU OpenBLAS and OpenScaLAPACK
- Make
- Python 2.6 or 2.7
- ARMCI (optional)
- Casper (optional)
- root or sudo permission (needed for installing program at global space)
NWChem documentation - Compiling NWChem
Look at the bottom of this page for example of Bash scripts
Execute the script to start compilation using the following commands
chmod +x INSTALL_NWCHEM_SCRIPT.sh
./INSTALL_NWCHEM_SCRIPT.sh
- Compilation and installation can take 20 - 60 min depending on configuration setting, compiler and system performance.
- If it is done successfully, the NWChem executable binary will be created at
$NWCHEM_TOP/bin/$NWCHEM_TARGET/nwchem
by default.
where $NWCHEM_TOP
and $NWCHEM_TARGET
are environment variables that define the top directory of nwchem-6.8.1
and platform of machine such as LINUX64
.
- Normal running
mpirun -np N /$NWCHEM_TOP/bin/$NWCHEM_TARGET/nwchem input.nw >& output.out &
- Normal running with specifying the number of processes per node
mpirun -np N -ppn M /$NWCHEM_TOP/bin/$NWCHEM_TARGET/nwchem input.nw >& output.out &
- Running with Casper
export NWCHEM_CASLIB=/full/path/to/casper/lib
mpirun -np N -ppn M -genv CSP_NG 1 \
-genv LD_PRELOAD $NWCHEM_CASLIB/libcasper.so \
/$NWCHEM_TOP/bin/$NWCHEM_TARGET/nwchem input.nw >& output.out
where N is the total number of MPI processes and M is the total number of MPI processes per node.
Note that the Casper is used for enhancing the performance of ARMCI method especially on cross-node MPI calculation.
- SGE
SGE scripts for submitting NWChem jobs
- SLURM
SLURM scripts for submitting NWChem jobs
- OpenMPI on Red Hat/CentOS
- Intel MPI on Red Hat/CentOS
- Intel MPI + GPU on Red Hat/CentOS
- MPICH MPI on Red Hat/CentOS
- MPICH MPI + GPU on Red Hat/CentOS
- MVAPICH2 MPI on Red Hat/CentOS
- MVAPICH2 MPI + GPU on Red Hat/CentOS
Rangsiman Ketkaew - [email protected]