基于GPU集群的科学计算可视化平台的性能测试与分析  

Scientific Visualization Application on GPU Cluster

作　　者：张亚南[1,2] 田东[1] 阚圣哲[1] 单桂华[1] 

机构地区：[1]中国科学院计算机网络信息中心,北京100190 [2]中国科学院研究生院,北京100190

出　　处：《科研信息化技术与应用》2011年第6期22-28,共7页E-science Technology & Application

基　　金：国家863计划资助项目(2011AA01A205)

摘　　要：应用于高性能计算集群的科学计算可视化软件发展日趋成熟,GPU加速的应用使得这类软件处理大规模数据的能力不断提高。本文介绍了可视化软件VisIt在GPU集群上安装及性能测试的一些工作,探讨了包括VisIt的基本架构、安装过程中遇到的问题及其解决方案、VisIt图像绘制中的并行性能测试、计算过程可视化的实现。The scientific visualization software in high-performance computing is more and more mature now and its capability dealing with massive data is highly increased by the acceleration of GPU cluster. In this paper, a successful port of Visit, which is a parallel, open source visualization and analysis tool, to a GPU cluster was discussed, including the architecture of Vislt, problems and solutions which were countered in the porting, the parallel performance of Visit in iso-surface, and the application of in-situ visualization.

关 键 词：可视化 VISIT GPU集群 计算过程可视化 科学计算可视化 

分 类 号：G252.7[文化科学—图书馆学]