The ProfugusMC proxy application targets a builtin problem geometry and provides command line control over the number of particles to simulate.
The problem geometry is a three dimensional core using a multigroup library with 252 groups. Vertically the core is broken down into a 3X3 grid with 2X2 fuel assembly in a corner. The fuel assembly is heterogenous with two UO2 assemblies and two MOX assemblies. The MOX assemblies use MOX 4.3%, MOX 7.0% and MOX 8.7%. The assemblies are described in a 17X17 grid. The full benchmark description is available at here (pdf), with the difference that ProfugusMC uses a 252 group cross-section library while the benchmark description uses 7 groups.
On ORNL’s Summit with the production code Shift, the ExaSMR typically runs 100-200k particles per GPU. Profugus is much cheaper per particle than Shift because of the difference in multigroup vs. continuous-energy cross sections so quite a few more particles per GPU are needed to get peak performance. Expect to run at least 1M particles per GPU for Profugus. An exascale simulation with ~30B particles is complete overkill for types of problems that Profugus is capable of running, but it’s perfectly valid for a computational exercise.
To run a problem on 4 GPUs with 4M particle, one would specify ./profugus -np 4000000