SPH Simulations with Reconfigurable Hardware Accelerator

Abstract

We present a novel approach to accelerate astrophysical hydrodynamical simulations. In astrophysical many-body simulations, GRAPE (GRAvity piPE) system has been widely used by many researchers. However, in the GRAPE systems, its function is completely fixed because specially developed LSI is used as a computing engine. Instead of using such LSI, we are developing a special purpose computing system using Field Programmable Gate Array (FPGA) chips as the computing engine. Together with our developed programming system, we have implemented computing pipelines for the Smoothed Particle Hydrodynamics (SPH) method on our PROGRAPE-3 system. The SPH pipelines running on PROGRAPE-3 system have the peak speed of 85 GFLOPS and in a realistic setup, the SPH calculation using one PROGRAPE-3 board is 5-10 times faster than the calculation on the host computer. Our results clearly shows for the first time that we can accelerate the speed of the SPH simulations of a simple astrophysical phenomena using considerable computing power offered by the hardware.