Global Memory Access Modelling for Efficient Implementation of the Lattice Boltzmann Method on Graphics Processing Units

International audienceIn this work, we investigate the global memory access mech- anism on recent GPUs. For the purpose of this study, we created spe- cific benchmark programs, which allowed us to explore the scheduling of global memory transactions. Thus, we formulate a model capable of estimating the execution time for a large class of applications. Our main goal is to facilitate optimisation of regular data-parallel applications on GPUs. As an example, we finally describe our CUDA implementations of LBM flow solvers on which our model was able to estimate performance with less than 5% relative error