| 基于CUDA架构矩阵乘法的研究 |
| |
| 引用本文： | 马梦琦,刘羽,曾胜田.基于CUDA架构矩阵乘法的研究[J].微型机与应用,2011,30(24):62-64,68. |
| |
| 作者姓名： | 马梦琦 刘羽 曾胜田 |
| |
| 作者单位： | 桂林理工大学信息科学与工程学院,广西桂林,541004 |
| |
| 摘 要： | 首先介绍了CUDA架构特点,在GPU上基于CUDA使用两种方法实现了矩阵乘法,并根据CUDA特有的软硬件架构对矩阵乘法进行了优化。然后计算GPU峰值比并进行了分析。实验结果表明,基于CUDA的矩阵乘法相对于CPU矩阵乘法获得了很高的加速比,最高加速比达到1079.64。GPU浮点运算能力得到有效利用,峰值比最高达到30.85%。
|
| 关 键 词： | CUDA 矩阵乘法 加速比 峰值比 |
Research of matrix multiplication based on CUDA architecture |
| |
| Ma Mengqi,LiuYu,Zeng Shengtian.Research of matrix multiplication based on CUDA architecture[J].Microcomputer & its Applications,2011,30(24):62-64,68. |
| |
| Authors: | Ma Mengqi LiuYu Zeng Shengtian |
| |
| Affiliation: | Ma Mengqi,LiuYu,Zeng Shengtian(School of Information Science and Engineering,Guilin University of Technology,Guilin 541004,China) |
| |
| Abstract: | This paper firstly introduced the characteristics of CUDA architecture, realized matrix multiplication using two ways on the GPU, and optimized the matrix multiplication according to unique hardware and software architecture based on CUDA. Then calculated and analyzed the peak ratio of GPU. Experimental results showed that CUDA-based matrix multiplication on the GPU achieved a higher speed-up ratio compared with that on tbe CPU. The maximum speedup to 1 079.64. The capability of floatingpoint calculations on the GPU was effectively taken advantage of,the highest peak ratio reached more than 30.85%. |
| |
| Keywords: | CUDA matrix multiplication speed-up ratio peak ratio |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
