For exascale-class systems this trend is expected to continue, with both the Aurora and Frontier exascale supercomputers slated to make heavy use of GPUs.Īdapting CPU-optimized codes for GPUs is not always straightforward. For instance, several current leadership-class computers (e.g., Summit and Sierra ) rely on GPUs to provide more than 95% of their floating-point throughput. In the field of high-performance computing, graphics processing unit (GPU) architectures are rapidly becoming the norm for supercomputers. For larger 3D simulation problems we found the new variant improved strong scaling by 3x when run on the Summit supercomputer. For very small 2D simulation problems we found the new immortal ray variant allowed for up to a 4.4x speedup when run on a single GPU.

The immortal ray technique exposes significantly more fine-grained parallelism by fundamentally reformulating the numerical details of ray discretization, resulting in performance tradeoffs with significant overall benefit on GPUs. In this study, we report on a newly developed “immortal ray” variant of TRRM. For smaller problems, or for large problems that are domain decomposed across many computational nodes, the problem size per node has insufficient parallelism to saturate GPU node resources, thus greatly limiting speedup. When porting to GPU-based systems, however, new performance challenges arise that are unique to processors targeting massive fine-grained parallelism. TRRM has demonstrated excellent performance on 3D nuclear reactor benchmark problems using CPU-based compute systems. The Random Ray Method (TRRM) is a recently developed adaptation of the Method of Characteristics for neutral particle transport simulations.

New immortal technique 2014