Non uniform LBM on GPGPUs

Non uniform LBM on GPGPUs

Second-order accurate mesh refinement for a D3Q27 Cumulant lattice Boltzmann method on GPGPUs

Researcher

Dipl.-Ing. Martin Schönherr

Computational efficiency is one of the main objectives in the field of Computational Fluid Dynamics (CFD). Multiple publications demonstrated that LBM implementations on General Purpose Graphics Processing Units (GPGPU) lead to a high computational performance when compared to implementations on CPUs.

We develop a GPGPU lattice Boltzmann implementation with three-dimensional hierarchical grid refinement and second-order accurate no-slip boundary conditions. With respect to the limited memory of the GPGPUs, we develop an implementation which requires only one set of distribution functions and thus saves almost half of the memory required in most GPGPU LBM implementations. Further GPGPU memory is saved by using an indirect addressing memory pattern. Our implementation is based on a D3Q27 cumulant lattice Boltzmann model which shows remarkably good results at high Reynolds number without any explicit turbulence model.

  • Optimize the cumulant lattice Boltzmann code
  • Grid refinement
  • Indirect Addressing