GPU-acceleration of Navier-Stokes solvers for compressible wall-bounded flows: the case of URANOS

The present paper describes the performance of URANOS, a high-fidelity Direct and Large- Eddy Simulation Navier-Stokes solver specifically developed for wall-bounded compressible flows. The code combines cutting-edge numerical methods peculiarly developed for high-speed turbulent flow simulations and is tailored to modern high-performance computing systems due to MPI parallelization combined with multi-GPUs communication access. In particular, OpenACC directives are implemented for GPU enabling offloading computational loads onto accelerators cards, making URANOS an easily maintained solver as well as guaranteeing extreme flexibility and portability. The solver validation is detailed for a broad range of Mach numbers, from low-speed to compressible cases. In particular, velocity statistics and Reynolds stress components for canonical channel flow and turbulent boundary layer configurations obtained with URANOS well agree with high-quality DNS data. Computational performance and scaling properties are tested on several multi-GPU-equipped clusters. Thus, with URANOS, the scientific community can take advantage of a GPU-accelerated solver in dealing with fluid modeling for aerodynamics applications. The source code is available under a BSD license at the following link: https://gitlab.com/fralusa/uranos_gpu