Massively-parallel multi-GPU simulations for fast and accurate automotive aerodynamics

In this paper, the scaling properties of the commercial Computational Fluid Dynamics (CFD) solver ultraFluidX are analyzed. The solver is based on the Lattice Boltzmann Method (LBM), is explicit in time, inherently transient and only requires next-neighbor information on the computational grid. ultraFluidX thus highly benefits from the computational power of Graphics Processing Units (GPUs). Additionally, the solver can make use of multiple GPUs on multiple compute nodes through an efficient implementation based on CUDA-aware MPI. The capabilities of the solver have already been demonstrated to various automotive OEMs around the world through several validation projects. This paper focuses on the scalability of ultraFluidX on single- and multi-node configurations in large-scale multi-GPU environments. Weak and strong scaling results for two selected test cases are reported and demonstrate that multi-GPU flow solvers have the potential to deliver high-fidelity results overnight. copyright © Crown copyright (2018).All right reserved.     «

In this paper, the scaling properties of the commercial Computational Fluid Dynamics (CFD) solver ultraFluidX are analyzed. The solver is based on the Lattice Boltzmann Method (LBM), is explicit in time, inherently transient and only requires next-neighbor information on the computational grid. ultraFluidX thus highly benefits from the computational power of Graphics Processing Units (GPUs). Additionally, the solver can make use of multiple GPUs on multiple compute nodes through an efficient imp...     »