摘要
基于交错网格的SIMPLE算法,利用CUDA(compute unified device architecture)技术进行了图形处理器(GPU)上的直接数值模拟(DNS).将高雷诺数方腔流作为研究实例,在NVIDIA GTX295显卡的单个和4个GPU上的计算速度最高可分别达Intel Xeon5430CPU之单核的50和150倍,与其他模拟结果的对比表明了上述计算的合理性,并展示了采用GPU实现高精度大规模的湍流计算的前景.
Based on SIMPLE arithmetic in a stageered grid system, Compute Unified Device Architecture (CUDA) was used to design and implement direct numerical simulation (DNS) on graphics processing units (GPU). High Reynolds number cavity flow was simulated on NVIDIA GTX 295, about 50 fold speedup of one GPU and 150 fold speedup of four GPUs over that of one core of the Intel Xeon 5430 CPU was achieved. The simulation results agreed with the literature data well. GPU accelerated DNS for high-accuracy and large-scale turbulent flow has a broad application prospect.
出处
《科学通报》
EI
CAS
CSCD
北大核心
2010年第20期1979-1986,共8页
Chinese Science Bulletin
基金
国家自然科学基金(20221603
20490201)
中国科学院知识创新工程重要方向性项目(KGCX2-YW-124)
国家重点基础研究发展计划(2009CB219906)资助
关键词
单相流动
直接数值模拟
CUDA
GPU
并行计算
single-phase flow, direct numerical simulation, CUDA, GPU, parallel computing