GPU平台上的叶轮机械CFD加速计算

通过数据并行的方式对一个成熟的叶轮机多块网格气动计算程序(MAP)进行了并行化处理,利用计算统一设备架构(CUDA)技术实现了在图形处理单元(GPU)上的并行计算.保留了原程序中的2阶空间迎风格式和隐式时间离散格式,并采用了隐式迭代对线性系统进行求解.经过2个叶轮机械算例的测试,与在传统的中央处理器(CPU)上运行的原程序相比,在计算结果完全一致的前提下,单GPU的计算速度最高可达单CPU计算速度的8.89倍,与四核并行的CPU计算相比可以得到2.39倍的加速.

Accelerated CFD computing of turbomachinery on GPU platform

The well-developed turbomachinery computational fluid dynamics code multi-block aerodynamic prediction (MAP) was solved by data-parallelized computing and implemented on a graphic processing unit(GPU) platform with the help of compute unified device architecture (CUDA) technology.The second-order upwind spatial scheme and the implicit temporal scheme in the original program were retained,while the linear system was solved by implicit iterations.During the test of two turbomachinery examples on a single GPU,a speed-up appeares to be 8.89 times comparing with one central processing unit (CPU) process and 2.39 times comparing with four CPU processes,with no extra deviation induced into the result.