Directable, high-resolution simulation of fire on the GPU

Abstract

The simulation of believable, photorealistic fire is difficult because fire is highly detailed, fast-moving, and turbulent. Traditional gridbased simulation models require large grids and long simulation times to capture even the coarsest levels of detail. In this paper, we propose a novel combination of coarse particle grid simulation with very fine, view-oriented refinement simulations performed on a GPU. We also propose a simple, GPU-based volume rendering scheme. The resulting images of fire produced by the proposed techniques are extremely detailed and can be integrated seamlessly into film-resolution images. Our refinement technique takes advantage of perceptive limitations and likely viewing behavior to split the refinement stage into separable, parallel tasks. Multiple independent GPUs are employed to rapidly refine final simulations for rendering, allowing for rapid artist turnaround time and very high resolutions. Directability is achieved by allowing virtually any user-defined particle behavior as an input to the initial coarse simulation. The physical criteria enforced by the coarse stage are minimal and could be easily implemented using any of the wide variety of commercially available fluid simulation tools. The GPU techniques utilized by our refinement stage are simple and widely available on even consumer-grade GPUs, lowering the overall implementation cost of the proposed system.