AsterX: a New Open-Source GPU-accelerated GRMHD Code for Dynamical Spacetimes

We present AsterX, a novel open-source, GPU-accelerated, fully generalrelativistic magnetohydrodynamic (GRMHD) code designed for dynamic spacetimesin 3D Cartesian coordinates, and tailored for exascale computing. We utilizeblock-structured adaptive mesh refinement (AMR) through CarpetX, the new driverfor the Einstein Toolkit, which is built on AMReX, a software framework formassively parallel applications. AsterX employs the Valencia formulation forGRMHD, coupled with the 'Z4c' formalism for spacetime evolution, whileincorporating high resolution shock capturing schemes to accurately handle thehydrodynamics. AsterX has undergone rigorous testing in both static and dynamicspacetime, demonstrating remarkable accuracy and agreement with other codes inliterature. Benchmarking the code through scaling tests on OLCF's Frontiersupercomputer, we demonstrate a weak scaling efficiency of about 674096 nodes compared to an 8-node performance.