You may download a PDF copy of this article from here.

This work was pursued to explore new combination of simulation method and computing hardware that will be capable of simulating air flow over large city models. Such simulations are normally thought as impractical since large city models tend contain many thousands of buildings of complex shapes. However, the current work was of interest to our customers from departments of environment, pollution and disaster management, who needed a tool to dynamically simulate the air flow over a city during normal or emergency scenarios.

Zeus Numerix chose to develop LBM for GPU based computing hardware. The LBM is a discrete computational technique where flow state at each lattice point is defined by a distribution function of fluid particles. These particles collide and propagate at each time step. The major advantage of LBM is that it can be readily parallelized and can handle complex geometries where traditional computational techniques may fail. Graphical Processing Unit (GPU) is specialized circuit designed to rapidly manipulate memory for computer graphics output. Their highly parallel structure makes them more efficient than CPUs, especially for algorithms like LBM.

Left: Flow over a model city block 1. Right: Flow over model city block 2.

LBM simulation of air flow over city blocks utilized 96 CUDA core on nVIDIA GT 440 GPU card. The run-time for 2 million lattice points was less than 15 minutes. Full scale city model simulation is underway on larger GPU clusters of our collaborator.