A Parallel AMR Method for Binary Black Hole Simulation

Binary Black Hole (BBH) simulations can lead to great scientific discovery but it is a challenging computationally intensive problem. The existing parallel algorithm and software cannot scale well on large supercomputers and use the latest hardware to accelerate the application. A new parallel algorithm is designed which can further improve the scalability of BBH simulations. Experimental results show that the existing method can only achieve 5X speedup at most (compared with the normalized the speed of 32 MPI processes); our scalable algorithm with GPU acceleration can achieve about 28X speedup. At the same time, our scalable GPU acceleration method can achieve almost ideal weak scaling up to 2048 computing nodes (GPUs) which guarantee that our method can really handle even larger BBH simulation efficiently.