Toward GPU-accelerated Traffic Simulation and Its Real-Time Challenge

REACTION 2012. 1st International workshop on Real-time and distributed computing in emerging applications. December 4th, 2012, San Juan, Puerto Rico.Traffic simulation is a growing domain of computational physics. Many life and industrial applications would benefit from traffic simulation to establish reliable transportation systems. A core challenge of this science research, however, is its unbounded scale of computation. This paper explores an advantage of using the graphics processing unit (GPU) for this computational challenge. We study two schemes of maximizing GPU performance in the context of traffic simulation, and provide some basic experiments. The experimental results show that our GPU implementation improves simulation speed by five times over the traditional CPU implementation. We also discuss that additional orders-ofmagnitude improvements could be achieved by overcoming the current hardware limitation of the GPU.This work was supported by Grant-in-Aid for Scientific Research 24500058