General Relativistic Magnetohydrodynamics on GPUs

As we enter the exascale era, numerical astrophysics requires codes that use GPUs efficiently to convert massive parallelism into science. In this talk, I will first outline the current landscape of GPU-enabled general-relativistic magnetohydrodynamics (GRMHD) codes and then introduce AsterX, an open-source code for dynamical spacetimes built on the CarpetX infrastructure, the new driver of the Einstein Toolkit. AsterX implements the flux-conservative Valencia formulation and evolves a staggered vector potential. It also leverages block-structured adaptive mesh refinement via the AMReX framework. AsterX has undergone rigorous testing on high-performance systems such as Frontera and Frontier, passing stringent GRMHD tests, which I will present along with latest benchmarks and applications to magnetized binary neutron star mergers as well as supermassive black-hole binary mergers. I will conclude with near-term developments, including expanded microphysics and support for moment-based neutrino transport.