Accelerating General Relativistic Radiation Magnetohydrodynamic Simulations with GPUs

The physical understanding of black hole accretion flow and outflow is directly related to understanding the energy production mechanisms in high-energy compact phenomena such as active galactic nuclei, X-ray binaries, and gamma-ray bursts, and also plays an important role in understanding the growth process of massive black holes. In the accretion disk surrounding a black hole, matter is trapped by the black hole’s strong gravity, and the energy is converted into heat, magnetic field, and radiation. The resulting radiation and magnetic fields accelerate some of the matter, forming jets and disk winds. To understand these complex phenomena, it is necessary to simultaneously solve the electromagnetic hydrodynamic equations and the radiation transfer equations in the strong gravitational field around the black hole, which involves a very large computational burden. With conventional CPU-based implementations, it is difficult to complete such large-scale calculations in a realistic time. Therefore, in this study, GPU offloading was introduced and radiation transfer equations and Magnetohydrodynamic calculations, which are particularly computationally demanding, were assigned to GPUs to accelerate the calculations. As a result, a speedup of about 10.7 times over the reference implementation with CPU-based flat MPI was achieved. This speedup is expected to enable more detailed reproduction of the dynamics of black hole accretion disks and jet streams, and to provide new insights into the energy production mechanisms of high-energy celestial objects.