Research On Acceleration Of PIC Method Based On GPU Cluster

“Fast ignitor” is a new technology of laser fusion, and the “Particle inCell” method is used widely recent years giving the credit to its accuracy andhigh cost-effective. In this research, we study about the particle in cellsimulation program of fast ignitor method. A parallel solution based onNvidia CUDA architecture has been designed, in order to accelerate the PICprogram. The parallel version of program has been executed on a workstationbased on heterogeneous computing architecture containing Intel CPU andNvidia GPU. The speedup of hotspot function has achieved more than10times, and the speedup of entire program has achieved about4times. In thisresearch, guided by the theory of load balancing, we have profiled theoriginal program in different ways in order to find out where are the hotspotsof the program. We have eliminated the dependency between the adjacentelements by building new attribute arrays to instead of the linked list. Afunction about atomic plus has been build in order to solve the problem ofsimultaneous writing. Besides, we have designed a “Transfer Array” forreducing the cost of data transfer. These contributions could be usd forreference in solving many other similer problems.