Abstract

Pre-compiled versions of LAMMPS available at CAVS.

Author(s): Mark A. Tschopp

LAMMPS Executables

This is how you access pre-compiled versions of LAMMPS at CAVS.

1. Logon to the cluster that you are going to be running your simulations on.
2. Type swsetup lammps. This sets up the paths to a directory that contains the LAMMPS binary files.
3. Type lmp_ and then hit TAB twice. This command will show the different LAMMPS versions that are available and already compiled.

lmp_12Oct10			lmp_21May08			lmp_4Jul10			lmp_exe
lmp_openmpi			lmp_XC
lmp_12Oct10-2			lmp_29Jul11			lmp_dxa
lmp_fftw3			lmp_SuSE10_verJan2010

4. These versions have different options and represent different versions. For example, lmp_fftw3 is the September 2011 version compiled with fftw3 for biological simulations. It is also compiled with most options that would be needed for simulations ran at HPC and CAVS. Please contact help desk or Mark Tschopp for adding additional packages.
5. Don't know which version the precompiled LAMMPS executable is? For example, what version of LAMMPS is lmp_exe? To check version quickly, type "lmp_exe", hit enter, and then hit Ctrl+C to stop the simulation. The LAMMPS version is printed to screen. The commands may change between different LAMMPS versions.

LAMMPS in Parallel on Unix

The following is an example of running in serial from the command line on UNIX systems where "input_script" is the input script and "lmp_exe" is the LAMMPS executable:

lmp_exe < input_script

To run this on four processors, simply add the following "mpirun" command in front of the script with the number of processors ("-np") equal to 4:

mpirun -np 4 lmp_exe < input_script

Now to run on 192 processors, this script is easily modified (but it runs a whole lot faster for large numbers of atoms):

mpirun -np 192 lmp_exe < input_script 

Beware: Large numbers of processors is not always faster. For small systems of atoms, the communication between processors may actually take a significant amount of time, which can cause it to run slower than less processors.