RadEnvCombined-ECPSprint | EXAALT and Kokkos: Making Exascale Simulations of Material Behavior a “SNAP”

Figure 2. MD simulation of helium and neutron damage in tungsten zirconium carbide, modeled using the SNAP machine learning potential in LAMMPS. Helium clusters form at the top surface of the polycrystalline tungsten. The simulation was run on OLCF Frontier using 46,656 AMD MI250X GCDs for approximately one (aggregate) Frontier-day. The scale of this representative microstructure was enabled by SNAP performance improvements. (Image credit: Mitch Wood, SNL).