基于图形处理器加速的叶轮机流场数值模拟研究  

An Accelerated Numerical Simulation Research on Flows in Turbomachines Based on Graphics Hardware

作　　者：张翔[1] 黄秀全[1] 

机构地区：[1]西北工业大学动力与能源学院,西安710072

出　　处：《科学技术与工程》2013年第11期3195-3199,共5页Science Technology and Engineering

摘　　要：近年来,图形处理器(GPU)已经逐渐发展成一种能够满足通用计算的多核心细粒度并行化的处理器,它往往能够提供10倍于CPU的浮点计算能力和更高的存储带宽,在其上开发计算流体力学(CFD)求解器正成为一种趋势。通过采用Jameson有限体积中心差分格式和四步Runge-Kutta时间推进法求解圆柱坐标系下的三维定常欧拉方程来模拟叶轮机械内部流场,并将原有运行在CPU上的代码移植到GPU上。通过比较,获得相同的流场计算结果;在运行速度上,获得了一个数量级的提升。In recent years ,the graphics processor（GPU） has gradually evolved into a multi-core fine-grained par- allel processor which can meet the general purpose and offer higher bandwidth and 10 times more floating point com- puting power than a modern CPU. CFD solver development based on GPU has been a trend. A fully 3D Euler method is applied to calculation of steady flows through turbomachines. The solver adopts the cell-centred finite volume dis- cretizaion and the 4-stage Runge-Kutta time-marching scheme in a cylindrical coordinate. GPU codes generate the same results as did on the CPU and speed up an order of magnitude by transplanting CPU code to run on the GPU.

关 键 词：GPU通用计算 CFD CUDA 叶轮机 

分 类 号：V231.3[航空宇航科学与技术—航空宇航推进理论与工程]