A new approach to the lattice Boltzmann method for graphics processing units

Abstract

International audienceEmerging many-core processors, like CUDA capable nVidia GPUs, are promising platforms for regular parallel algorithms such as the Lattice Boltzmann Method (LBM). Since the global memory for graphic devices shows high latency and LBM is data intensive, the memory access pattern is an important issue for achieving good performances. Whenever possible, global memory loads and stores should be coalescent and aligned, but the propagation phase in LBM can lead to frequent misaligned memory accesses. Most previous CUDA implementations of 3D LBM addressed this problem by using low latency on chip shared memory. Instead of this, our CUDA implementation of LBM follows carefully chosen data transfer schemes in global memory. For the 3D lid-driven cavity test case, we obtained up to 86% of the global memory maximal throughput on nVidia's GT200. We show that as a consequence highly efficient implementations of LBM on GPUs are possible, even for complex models

Scalable lattice Boltzmann solvers for CUDA GPU clusters

“…As shown in [12], the misalignment overhead is significantly higher for store operations than for read operations. We therefore suggested in [13] to use the in-place propagation scheme outlined by Fig. 3 instead of the ordinary out-of-place propagation scheme illustrated in Fig.…”

The TheLMA project: A thermal lattice Boltzmann solver for the GPU

“…Figures 4 and 5 outline the two propagation schemes (in the two-dimensional case, for the sake of simplicity). It was shown in [17] that the cost of misaligned reads is of the same order of magnitude than the overhead of a rearrange kernel. It should be noted that the in-place propagation approach is simpler and exerts less pressure on hardware than the shared memory approach.…”

“…Our implementation is based on the isothermal flow solver described in [17]. The lattice is a rectangular cuboid of dimensions N x × N y × N z .…”

“…Using the least significant dimension possible to span the velocity distribution reduces the occurrences of TLB misses. We experimented a 13% performance improvement over the major dimension version which is used in [17].…”

“…This program is an extended and improved version of the isothermal solver decribed in [17]. It is part of the TheLMA project [19] which aims at providing a comprehensive framework for implementing LBM solvers on GPUs and other emerging many-core architectures.…”