CPU+多GPU异构协同计算的体系结构研究

Research of Architecture of CPU + Multi-GPU Heterogeneous Collaborative Computing

以CUDA架构为例,对传统的CPU+单GPU架构进行了分析,提出了一种CPU+多GPU异构协同计算的系统方案,对关键的CPU对多GPU的管理及多GPU间数据通信等问题做了重点讨论,从理论上进行了可行性分析,并提出了相应的优化方法.

In this paper ,we analyzed the traditional CPU ＋ single GPU architecture ,and put forward a CPU ＋multi-GPU heterogeneous collaborative computing system .We focused on the CPU management for multi-GPU and data communication between multi-GPU ,then theoretically analyzed the feasibility ,finally gave the corresponding optimization methods .