GROMACS
Summary
The source code for Gromacs can be downloaded from ftp://ftp.gromacs.org/pub/gromacs/gromacs-4.5.5.tar.gz. The compilation instructions is found in the source. The input files can be downloaded from ftp://ftp.gromacs.org/pub/benchmarks/gmxbench-3.0.tar.gz
Instructions
This benchmark will use the input in d.dppc/ only.
Running the benchmark:
# tar zxf gmxbench-3.0.tar.gz # cd d.dppc # sed -i ‘/nsteps/ s/5000/150000/’ grompp.mdp # grompp -v # time mpirun -n 256 /path/to/gromacs/bin/mdrun_mpi
The walltime in seconds as measured by the time command for a 256 CPU-core run is used in the calculation of benchmark performance.
Comment: The original d.dppc benchmark performs 5000 time steps. This is too few to get realistic benchmark numbers and one has to increase the number of time steps to 150000 steps. The vendor is allowed to scale the number of time steps up even further if it feels that the run is too short to create realistic benchmark numbers. The timings reported should then be scaled back down to the time for 150000 time steps. Any such scaling of time steps must be documented in detail.
The benchmark report must include detailed runtime description about
- compiler version
- math libraries used
- system configuration
- compiler options used to build the application
- how many cores used per compute node
- All output and output files generated by the execution.
Verified output
Verified output results can be found here.

