Lattice Boltzmann Method Simulations on Massively Parallel Multi-core Architectures

We describe the implementation and optimization of a state-of-the-art Lattice Boltzmann code for computational fluid-dynamics for massively parallel systems using multi-core processors. We consider a code describing 2D compressible fluid flows, including thermal and combustion effects. We carefully match and balance the large degree of parallelism of the underlying algorithm with all available parallel resources (inter-node, intra-node, SIMD). We test our code on the prototype of the application-driven AuroraScience machine; our results can be readily ported to virtually any large scale system. We obtain a sustained performance for this ready-for-physics code that is a large fraction of peak performance.