A Fast Hirbulence Generator using Graphics Processing Units

摘要

We present a fast turbulence generator using Graphics Processing Units (GPUs) that enables us to generate a turbulence field by factors ranging from 4-to-7S times faster compared to using Central Processing Units (CPUs). Our turbulence generator is based Kraichnan's classic technique of generating synthetic isotropic turbulence and will be deployed online for access by the aeroacoustics, aerospace, and fluid dynamics communities. Our approach takes an arbitrary energy spectrum as input and computes a velocity vector field that reproduces the input spectrum up to the Nyquist limit. The Nyquist limit is dictated by the domain size and grid resolution specified by the user. The velocity field is stored on a staggered grid and satisfies the discrete divergence-free requirement when using periodic boundary conditions. The turbulence generator is written in Python using the numba library and is freely available to download from https: //github. com/saadgroup. The errors produced by the generator are examined for canonical spectra such as the Comte-Bellot-Corrsin (CBC) and the von Karman-Pao (VKP) spectra. The scalability of our code is verified at various grid resolutions. The synthetic turbulence is suitable for a priori and a posteriori verification of turbulence codes and models as well as noise-generation models for aeroacoustic calculations.