Interactive Fluid Simulation On The GPU

Fluid simulation has been successfully used in the computer graphicscommunity to generate detailed and physically-plausible motion of fluids. Avariety of effective and stable methods have been developed forhigh-precision simulation of fluid phenomena. However, due to the highcomputational overhead，these methods have been only used for off-lineapplications like movie effects. Creating and rendering realistic fluids atinteractive frame rates remains a challenging and interesting problem.We present a fluid dynamic system, which allows to moving objects andreal-time simulations of large scale three dimensional smokes. The system ishigh in precision and visual effects.The core of the system is an Eulerian fluid solver which is well suited forreal-time application since it is unconditionally stable, takes constantcalculation time per frame, and with high computing precision. We introducea specialized geometry multigrid algorithm combined with variationalframework for solving the Poisson equation. Moreover, the solver can run onGPUs and make full use of its parallel computer resources. Experimentalresults show that our solver has significant performance advantages comparedwith other solvers. It can meet the need of real-time application.We implement a GPU-accelerated volume rendering system that treats theparticles as participating media and can render effects such as smoke, dust, ormist. The rendering method has the advantages of easy programing, highvisual effects and quick calculating speed. With the help of GPU, we canspeed up the sorting of the particles, and avoid loading redundant data andtherefore to reduces the bandwidth requirement. To further improve the performance of the fluid simulation system, we useGPU technologies such as ’texture memory’,’coalescent pattern’ to exploit thecomputing power of Nvidia GPUs. These technologies accelerate some partsof the system to varying degrees.This combination of related algorithms leads to real-time simulationsystem. We show the effectiveness of the proposed methods in severalscenarios, all running at interactive rates on a modern GPU. We can imaginethat our approach will run faster and faster in the feature due to hardwareimprovement and will possess important applications in many fields.