通用图形处理器GPGPU的并行计算研究  

Research on Parallel Computing of General Purpose CPGPU

作　　者：张鹏博[1] 郭兵[1] 黄义纯 曹亚波[1] 

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《单片机与嵌入式系统应用》2017年第8期40-43,共4页Microcontrollers & Embedded Systems

基　　金：国家自然科学基金重点项目(项目编号:61332001);国家自然科学基金资助项目(项目编号:61272104)

摘　　要：随着图形处理器(GPU)从仅用来进行图形图像渲染,脱离成为并行计算平台通用图形处理器(GPGPU),其计算能力越来越强,本文在研究GPGPU体系结构的基础上对GPGPU并行计算线程调度进行深入研究,阐述了GPU线程调度原理,揭示了SIMT调度模式的不足。通过公式推导阐述了系统功耗与系统运行频率的关系。The Graphics Processing Unit （GPU） only is been used to the graphics image rendering, so it becomes a parallel computing platform General Purpose GPU（GPGPU）,the parallel computing power is growing, and the application is more and more widely.Based on the study of GPGPU architecture, the GPGPU parallel computing thread scheduling is studies,and the principle of GPU thread sched- uling and the deficiency of SIMT scheduling mode are introduced.The relationship between system power consumption and system oper- ating frequency is expounded by formula.

关 键 词：大数据 并行计算 线程调度 GPU节能 

分 类 号：TP332[自动化与计算机技术—计算机系统结构]