Large-scale wind simulation using lattice Boltzmann method on GPU-Rich supercomputer

Onodera, Naoyuki   

Recently, GPU has been emerging as one of high performance devices to realize a large-scale simulation with less power consumption. We realize a high precision wind simulation based on the LBM. Our code is tuned to achieve high performance on the Pascal GPU architecture, and we propose a communication reduced algorithm with a temporal blocking method. We performed the strong scalability measurements with TSUBAME3.0 which has 4 GPUs and 2 CPU sockets per node. Regarding the strong scalability results, the performance of GPU with a single node is 30 times higher than that of CPU, and the performance of GPU with 36 nodes is 10 times higher than that of CPU.