基于CPU/GPU集群的编程的研究

Research on Programming Based on CPU/GPU Heterogeneous Computing Cluster

随着微处理器技术的发展,GPU/CPU的混合计算已经成为是科学计算的主流趋势.本文从编程的层面,介绍了如何利用已有的并行编程语言来,调度GPU的计算功能,主要以MPI(一种消息传递编程模型)与基于GPU的CUDA(统一计算设备架构)编程模型相结合的方式进行GPU集群程序的测试,并分析了CPU/GPU集群并行环境下的运行特点.从分析的特点中总结出GPU集群较优策略,从而为提高CPU/GPU并行程序性能提供科学依据.

With the fast development in computer and microprocessor, Scientific Computing using CPU/GPU hybrid computing cluster has become a tendency. In this paper, from programming point of view, we propose the method of GPU scheduling to improve calculation efficiency. The main methods are through the combination of MPI （Message Passing Interface） and CUDA （Compute Unified Device Architecture） based on GPU to program. According to running condition of the parallel program, the characteristic of CPU/GPU hybrid computing cluster is analyzed. From the characteristic, the optimization strategy of parallel programs is found. So, the strategy will provide basis for improving the CPU/GPU parallel program