Off-Loading LET Generation to PEACH2

Abstract

A hardware local essential tree (LET) generator used in an N-body simulation is implemented on the FPGA of PEACH2 (PCI Express Adaptive Communication Hub ver2), a low latency switching hub for high performance GPU clusters. By using the pipelined on-the-fly execution with a multipole acceptance criterion judging module and a data updating module, the generation performance is 2.2 times faster than that with the CPU. When data communication is considered, the performance was 7.2 times as the case with the CPU.