计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (33): 67-69.DOI: 10.3778/j.issn.1002-8331.2009.33.021

• 研发、设计、测试 • 上一篇    下一篇

图形处理器通用计算的实现与验证

齐 记1,杨孔庆1,杨 磊2   

  1. 1.集美大学 理学院复杂系统研究中心,福建 厦门 361021
    2.中国科学院 近代物理研究所,兰州 730000
  • 收稿日期:2008-09-19 修回日期:2008-12-23 出版日期:2009-11-21 发布日期:2009-11-21
  • 通讯作者: 齐 记

Enablement and verification of GPGPU

QI Ji1,YANG Kong-qing1,YANG Lei2   

  1. 1.Center for Complex System College of Science,Jimei University,Xiamen,Fujian 361021,China
    2.Institute of Modern Physics,Chinese Academy of Sciences,Lanzhou 730000,China
  • Received:2008-09-19 Revised:2008-12-23 Online:2009-11-21 Published:2009-11-21
  • Contact: QI Ji

摘要: 讨论了显示卡用于通用科学计算的问题,并以大型矩阵的基本运算问题详细比较了CPU和GPU计算之间的差别。在基本的矩阵运算中,运用适当的矩阵分块,GPU的计算速度比CPU快50倍左右。而且,显示卡低廉的价格为更多科研工作者实现大规模运算提供了可能。

关键词: 图形处理器(GPU), 图形处理器通用计算技术(GPGPU), 中央处理器(CPU)

Abstract: In this paper,the general purpose computing on Graphics Processing Unit(GPU) is discussed,and the fundamental operations of large scale matrice by GPU are studied in detail to compare the difference between the calculation of CPU and GPU.In the fundamental operations of large scale matrice,computing speed of GPU can be 50 times faster than CPU on the suitable matrix blocking.Moreover,the cheapness of the graphic card makes it possible that more scientific workers can do large-sized calculation.

Key words: Graphics Processing Unit(GPU), General Purpose Graphic Process Unit(GPGPU), Central Processing Unit(CPU)

中图分类号: