基于NUMECA FINE/Turbo的并行计算测试  

Parallel computing test based on NUMECA FINE / Turbo

作　　者：董晶[1] 

机构地区：[1]沈阳鼓风机集团股份有限公司,沈阳110869

出　　处：《计算机辅助工程》2014年第6期108-110,112,共4页Computer Aided Engineering

摘　　要：为具体了解CFD软件NUMECA FINE/Turbo的并行计算性能,良好把握后续的科研工作进度,分别研究在激活超线程情况下单节点计算与多节点并行计算以及CPU在激活超线程前、后计算速度的差异.结果表明:在多节点并行计算时,计算速度与实际参加并行计算的CPU物理核心数量成正比;在激活超线程的情况下,并行计算节点数在超过实际物理核心数后明显降低计算速度的提升.To understand the parallel computing performance of CFD software NUMECA concretely and have a good grasp of the work progress of the follow-up research, the single node computing and the multi-node parallel computing are separately researched in the case of activated hyper-thread, and the difference of computation speed of the CPU before and after activating hyper-thread is obtained. The results show that, the computation speed is proportional to the actual computational number of physical CPU cores in the ease of multi-node parallel computing; in the case that the hyper-thread is activated, the ascension of computation speed is obviously reduced when the number of parallel computing nodes is more than physical cores.

关 键 词：并行计算 超线程 计算节点数 CPU核心数量 计算速度 NUMECA FINE/Turbo 

分 类 号：TP338.6[自动化与计算机技术—计算机系统结构]