CP2K

CP2K is a quantum chemistry and solid state physics software package that can perform atomistic simulations of solid state, liquid, molecular, periodic, material, crystal, and biological systems.

Example job

Here we will use the static calculation example from the CP2K website

First, log into one of the Compute Canada clusters and download the needed files with the following commands:

wget https://www.cp2k.org/_media/static_calculation.tgz
tar xvfz static_calculation.tgz	
cd static_calculation/sample_output_no_smearing

Then, in that directory, create the following job submission script, with the account name changed to the one you are using.

File : mpi_job.sh

#!/bin/bash
#SBATCH --account=def-someuser
#SBATCH --ntasks=4               # number of MPI processes
#SBATCH --mem-per-cpu=4G      # memory; default unit is megabytes
#SBATCH --time=0-00:15           # time (DD-HH:MM)

module load cp2k/8.2
srun cp2k.popt -o Si_bulk8.out Si_bulk8.inp

To submit this job, execute:

sbatch mpi_job.sh

To see if the job completed, run the command

sq

If your job is no longer listed, that means it has completed.

The output of CP2K will be located in the file Si_bulk8.out. There will also be an output file named slurm-*.out which should be empty if the calculation completed without error.

Threaded/MPI jobs

The installation of CP2K version 8.2 and later includes both the MPI executable cp2k.popt and the OpenMP/MPI executable cp2k.psmp, which may give better performance for some calculations. Our test shows a 10% performance increase for QS/H2O-512.inp benchmark when using 2 threads per MPI process, compared to running MPI-only executable cp2k.popt (both runs used the same number of CPU cores in total).

Below is an example OpenMP/MPI job submission file for the Beluga cluster. To use on other clusters, the number of tasks would have to be adjusted to match the number of cores available on the nodes of that cluster. The performance change when threads are used is highly problem dependent, and running the cp2k.psmp executable may actually be slower for some cases, so you must benchmark your code and choose the right option for your particular case.

File : openmp_mpi_job.sh

#!/bin/bash
#SBATCH --account=def-someuser
#SBATCH --nodes=2
#SBATCH --ntasks=40               # number of MPI processes
#SBATCH --cpus-per-task=2
#SBATCH --mem-per-cpu=3G      # memory; default unit is megabytes
#SBATCH --time=0-00:59           # time (DD-HH:MM)

module load cp2k/8.2

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK
srun cp2k.psmp -o H2O-512.out H2O-512.inp

Message Passing Interface