论文部分内容阅读
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 evolved into a multi-core, fine-grained and parallelized processor that meets general-purpose computing. It often provides 10 times the floating-point computing power of the CPU and higher storage bandwidth. It is becoming a trend to develop computational fluid dynamics (CFD) solvers. The internal flow field of the impeller is simulated by using the Jameson finite volume center difference scheme and the four-step Runge-Kutta time-propulsion method to solve the three-dimensional steady Euler equation in the cylindrical coordinate system. The original code running on the CPU is ported to the GPU . By comparison, the same flow field calculation results are obtained; an order of magnitude increase is achieved at operating speed.