Accelerated CFD Simulations Using Eulerian and Lagrangian Methods on GPUs

An efficient aerodynamic design of vehicles is one possibility for the automotive industry to meet the increasingly restrictive emission regulations and also customer requirements regarding fuel economy. CFD methods have become a valuable tool to address such topics. However, todays detailed vehicle simulations require a considerable amount of time and hardware resources. Due to a massively parallel execution of numerical algorithms on GPUs, GPU computing offers significant reduction of turn-around times and has become an powerful alternative to CPU-based systems in the HPC sector. This paper presents recent developments for accelerating the widespread Navier-Stokes-based CFD methods by using a hybrid CPU-GPU approach. In addition, results of a Lattice-Boltzmann method, which has been implemented completely in CUDA for highly efficient computations on GPUs are shown.