Lattice Boltzmann methods on the ClearSpeed Advance™ accelerator board
Institute for Applied and Numerical Mathematics, Karlsruhe Institute of Technology (KIT),
Universität Karlsruhe (TH), Karlsruhe, Germany
2 SRG New Frontiers in High Performance Computing Exploiting Multicore and Coprocessor Technology, Karlsruhe Institute of Technology (KIT), Universität Karlsruhe (TH), Karlsruhe, Germany
Corresponding author: firstname.lastname@example.org
Numerical analysts and programmers are currently facing a conceptual change in processor technology. Multicore concepts, coprocessors and accelerators are becoming a vital part in scientific computing. The new hardware techno- logies lead to new paradigms and require adapted methodologies and techniques in numerical simulation. These developments play an important role in computational fluid dynamics (CFD) where many highly CPU-time demanding problems arise. In this paper, we propose a parallel lattice Boltzmann method (LBM) in the context of a coprocessor technology, the ClearSpeed Advance™ accelerator board. Implementations of LBMs on parallel architectures benefit from localities of the necessary interactions and the regular structure of the underlying meshes. The considered board supports high-level parallelism and double precision conforming to the IEEE 754 standard. However, the solution process relies on a huge amount of data which needs to propagate along the mesh. This prototypical fact shows up the bottleneck of internal communication bandwidth and indicates the limits of this type of small-scale parallel systems.
© EDP Sciences, Springer-Verlag, 2009