High order accurate simulation of compressible flows on GPU clusters over Software Distributed Shared Memory

摘要: Abstract The advent of multicore processors during the past decade and especially recent introduction many-core Graphics Processing Units (GPUs) open new horizons to large-scale, high-resolution simulations for a broad range scientific fields. Residing at forefront advancements in multiprocessor technology, GPUs are often chosen as co-processors when intensive parts applications need be computed. Among various domains, area Computational Fluid Dynamics (CFD) is potential candidate that could significantly benefit from utilization GPUs. In order investigate this possibility, we herein evaluate performance high accurate method simulation compressible flows. Targeting computer systems with multiple GPUs, current implementation respective evaluation taking place on GPU cluster. With respect using these paper offers an alternative mainstream approach message passing by considering shared memory abstraction. implementations presented paper, updates data not explicitly coded programmer across phases, but propagated through Software Distributed Shared Memory (SDSM). This way, intend preserve unified view extends hierarchy node level cluster level. Such extension facilitate porting multithreaded codes clusters. Our results indicate competitive paradigm they lay grounds further research use abstraction future