Performance-portable binary neutron star mergers with AthenaK

Thanks to their impressive floating-point performance, modern supercomputing facilities derive a significant portion of their computational power through graphical processing units (GPUs). Next-generation numerical relativity frameworks must be able to take advantage of these resources to tackle the problems currently facing the field. However, the broad availability of CPU-based machines (whether they be supercomputers or personal workstations) makes it desirable to have a single code which can run on both CPU-based and GPU-based machines. AthenaK is a new astrophysics code which achieves this goal through the Kokkos performance-portability framework. In this talk, I will focus on AthenaK's newly-developed numerical relativity capabilities, particularly its general-relativistic magnetohydrodynamics solver and applications to binary neutron star mergers. I will also show results highlighting the code's excellent performance and scaling on both CPUs and GPUs.