摘要
从2004年开始,图形处理器GPU的通用计算成为一个新研究热点,此后GPGPU(Gen-eral-Purpose Graphics Processing Unit)在最近几年中取得长足发展。从介绍GPGPU硬件体系结构的改变和软件技术的发展开始,阐述GPGPU主要应用领域中的研究成果及最新发展。针对各种应用领域中计算数据大规模增加的趋势,出现单个GPU计算节点无法克服的硬件限制问题,为解决该问题出现多GPU计算和GPU集群的解决方案。详细地讨论通用计算GPU集群的研究进展和应用技术,包括GPU集群硬件异构性的问题和软件框架的三个研究趋势,对几种典型的软件框架Glift、Zippy、CUDASA的特性和缺点进行较详细的分析。最后,总结GPU通用计算研究发展中存在的问题和未来的挑战。
The general purpose computation of graphic processing unit became a new research field since 2004. GPGPU has been developing rapidly in recent years at a high speed. Starting from an introduction to the development of the architecture of GPU for general-purpose computation and software technology, the study and development of GPU for general-purpose computation are introduced. Aiming at the large scale data of various application fields, GPU cluster is proposed to overcome the limitation of single GPU. So the development and application tech- nologies of GPGPU cluster are discussed and include the issue of heterogeneous cluster and the trend of software for GPU cluster. Several frameworks for GPU cluster are analyzed in detailed, such as Glift, Zippy, and CUDASA. Finally, the unsolved problems and the new challenge in this subject are proposed.
出处
《黑龙江大学自然科学学报》
CAS
北大核心
2012年第5期672-679,共8页
Journal of Natural Science of Heilongjiang University
基金
国家自然科学基金资助项目(60970012)
上海市信息技术领域重点科技攻关项目(09511501000)
上海市重点科技项目(09220502800)
上海市重点学科建设项目(S30501)
上海市教委基金资助项目(09YZ428
B08056)