A Further Development of the Asynchronous GPU CABARET Method for Jet-Noise Modelling

A new asynchronous modification of the CABARET method is proposed for the solution of Navier–Stokes equations in the Large Eddy Simulation regime. The modification is based on improvement of the asynchronous extrapolation step both for Euler and Navier–Stokes CABARET solver. The algorithm is implemented as a parallel code for NVIDIA GPU using multiple CUDA-cores with MPI multi-CPU support. The algorithm accelerated on Graphics Processing Units (GPUs) is applied for the jet flow simulations in the Wall Model Large Eddy Simulation framework. The efficiency of code parallelization is discussed. The suggested asynchronous CABARET algorithm provides an almost 5000 times acceleration of calculations compared to a single CPU core, and allows us to calculate 300 convective times of jet development per day on a grid of 16 million cells. The flow solutions are analysed and compared with TsAGI anechoic chamber experimental data. It is shown that the structure of the jet flow is reproduced correctly, capturing low-amplitude instability waves in the jet potential core and fine-scale turbulent fluctuations in the near-field. Far-field noise predictions in the Ffowcs Williams–Hawking formulation with the azimuthal decomposition of the far-field radiation reproduce the nontrivial spectra and directivities of individual far-field acoustic modes.