The Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) is a classical molecular dynamics code designed for high-performance simulation of large molecular systems. LAMMPS generally scales well on OSC platforms, offers a variety of modelling techniques, and offers GPU enhanced computation.
LAMMPS is available to all OSC users without restriction.
The following versions of LAMMPS are available on OSC systems:
To use LAMMPS, run the following command to first set up your environment:
module load lammps
To see other available versions, run the following command:
To see information on the packages and executables for a particular installation, run the module help command, for example on Oakley:
module help lammps/01Feb14
Once a module is loaded, LAMMPS can be run with the following command:
lammps < input.file
Sample batch scripts and LAMMPS input files are available here:
Below is a sample batch script for the Glenn Cluster. It asks for 16 processors and 10 hours of walltime. If the job goes beyond 10 hours, the job would be terminated.
#PBS -N chain #PBS -l nodes=2:ppn=8 #PBS -l walltime=10:00:00 #PBS -S /bin/bash #PBS -j oe module load lammps cd $PBS_O_WORKDIR pbsdcp chain.in $TMPDIR cd $TMPDIR lammps < chain.in pbsdcp -g '*' $PBS_O_WORKDIR
On the Glenn and Oakley OSC clusters, LAMMPS can run on GPU's. See the sample scripts for details. The following text is specific to the Glenn cluster and demonstrates using GPU's to speed up certain pair_style's. This example shows how to load and run such a GPU-enabled version. Here is a sample PBS script to run. It uses one node and two GPU's for the computation.
#PBS -N lammpsTest #PBS -l nodes=1:ppn=8,feature=gpu #PBS -l walltime=00:10:00 #PBS -S /bin/bash #PBS -j oe module switch mvapich2-1.4-gnu module load cuda-3.0 module load fftw2-2.1.5-double-mvapich2-1.4-gnu module load lammps-25Mar11cuda cd $PBS_O_WORKDIR cp lj-gpu.in $TMPDIR cd $TMPDIR mpiexec -np 2 lmp_osc < lj-gpu.in > out cp $TMPDIR/* $PBS_O_WORKDIR
Here is a sample input with the necessary modifications to use a GPU pair_style, and it uses both GPU's.
newton off units lj atom_style atomic lattice fcc 0.8442 region box block 0 20 0 20 0 20 create_box 1 box create_atoms 1 box mass 1 1.0 velocity all create 1.44 87287 loop geom pair_style lj/cut/gpu 2.5 pair_coeff 1 1 1.0 1.0 2.5 neighbor 0.3 bin neigh_modify delay 0 every 20 check no fix 0 all gpu force/neigh 0 1 1.0 fix 1 all nve timestep 0.003 thermo 100 run 100
Please note that the you cannot run more than two threads per node. More than two threads will cause the application to hang, since there are only two GPU's per node. The LAMMPS input and the pbs script must match up with the number of GPU's being used.
Please refer to the documentation for more pair_style's that can be used in such simulations.