Multi-GPU thermal lattice Boltzmann simulations using OpenACC and MPI

We assess the performance of hybrid Open Accelerator (OpenACC) and Message Passing Interface (MPI) approach for multi-graphics processing units (GPUs) accelerated thermal lattice Boltzmann (LB) simulation. The OpenACC accelerates computation on a single GPU, MPI synchronizes information between multiple GPUs. With two-dimension (2D) simulation achieved 1.93 billion updates per second (GLUPS) with grid number $8193^{2}$, three-dimension (3D) 1.04 GLUPS $385^{3}$, which is more than 76% theoretical maximum performance. On multi-GPUs, we adopt block partitioning, overlapping communications computations, concurrent to optimize parallel efficiency. show that in strong scaling test, using 16 GPUs, 2D 30.42 3D 14.52 GLUPS. In weak efficiency remains above 99% up Our results demonstrated that, improved data task management, technique promising LB multi-GPUs.