H-AMR: A New GPU-accelerated GRMHD Code for Exascale Computing With 3D Adaptive Mesh Refinement and Local Adaptive Time-stepping

General-relativistic magnetohydrodynamic (GRMHD) simulations have revolutionized our understanding of black hole accretion. Here, we present a graphics processing unit (GPU) accelerated GRMHD code \hammer{} with multi-faceted optimizations that, collectively, accelerate computation by 2-5 orders of magnitude for a wide range of applications. Firstly, it introduces a spherical grid with 3D adaptive mesh refinement that operates in each of the 3 dimensions independently. This allows us to circumvent the Courant condition near the polar singularity, which otherwise cripples high-resolution computational performance. Secondly, we demonstrate that local adaptive time-stepping (LAT) on a logarithmic spherical-polar grid accelerates computation by a factor of $\lesssim10$ compared to traditional hierarchical time-stepping approaches. Jointly, these unique features lead to an effective speed of $\sim10^9$ zone-cycles-per-second-per-node on 5,400 NVIDIA V100 GPUs (i.e., 900 nodes of the OLCF Summit supercomputer). We illustrate \hammer{}'s computational performance by presenting the first GRMHD simulation of a tilted thin accretion disk threaded by a toroidal magnetic field around a rapidly spinning black hole. With an effective resolution of $13$,$440\times4$,$608\times8$,$092$ cells, and a total of $\lesssim22$ billion cells and $\sim0.65\times10^8$ timesteps, it is among the largest astrophysical simulations ever performed. We find that frame-dragging by the black hole tears up the disk into two independently precessing sub-disks. The innermost sub-disk rotation axis intermittently aligns with the black hole spin, demonstrating for the first time that such long-sought alignment is possible in the absence of large-scale poloidal magnetic fields.