Oral Presentation

GAMER-2: GPU-accelerated Adaptive Mesh Refinement

Author(s): Hsi-Yu Schive (NTU)

Presenter: Hsi-Yu Schive (Institute of Astrophysics, National Taiwan University)

GAMER-2 is a multi-GPU-accelerated adaptive mesh refinement (AMR) code for astrophysics. It supports a rich set of physics and functionality, including hydro, MHD, wave dark matter, particles, chemistry and radiative processes, star formation, adaptive time-stepping, memory pool, bitwise reproducibility, load balancing, hybrid OpenMP/MPI/GPU, and data analysis with yt. To demonstrate its capability, we directly compare with Enzo on isolated disk galaxy simulations and with FLASH on galaxy cluster merger simulations, and show that GAMER not only produces very consistent physical results but also outperforms these popular AMR codes in strong scaling by one to two orders of magnitude. In addition, we demonstrate good weak scaling with a parallel efficiency of 60 – 70% and an overall performance of 5e10 cells/sec using 4,096 GPUs and 65,536 CPU cores simultaneously. GAMER-2 is open source, available at https://github.com/gamer-project/gamer.