Source Term-Based Turbulent Flow Simulation on GPU with Link-Wise Artificial Compressibility Method

Abstract

We present a GPU-based turbulent flow simulation by link-wise artificial compressibility method (LW-ACM). The standard implementations of the lattice Boltzmann method are limited by memory requirements due to the nature of the distribution functions employed. LW-ACM avoids the need to store the density distribution function via the use of a hybrid of LBM and finite difference method. This method, previously used only for simple cases without inlet/outlet boundary conditions, is here extended for general-purpose 3D turbulent flow via the introduction of the synthetic eddy method (SEM) as a distributed source term into the channel. A channel flow is performed to validate the implementation in this paper. Experimental results demonstrate performance on a single GPU of up to 11237 MLUPS and 4656 MLUPS in single and double precision, respectively, amongst the fastest results reported to date, demonstrating the practical opportunities this approach can offer for systematic evaluation of complex turbulent flow.

Keywords:

• Link-wise artificial compressibility method
• synthetic eddy method
• turbulent flow simulation
• lattice Boltzmann method
• boundary condition
• computational fluid dynamics
• GPU

Acknowledgments

We would like to acknowledge the assistance given by Jiexin Zhang, Adrian Harwood, Andrew Mole, Ruizhi Liu, Chengkun Wu, Research IT and the use of the Computational Shared Facility at The University of Manchester.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work is funded by Tianhe Supercomputer Project under grant no. 2018YFB0204301, the National Natural Science Foundation of China under grant no. 61902413, and the open funds from the State Key Laboratory of High-Performance Computing under grant nos. 201901-12 and 202001-04.