Tamás Umenhoffer and László Szirmay-Kalos
Department of Control Engineering and Information Technology, Technical University of Budapest,
Budapest, Magyar tudósok krt. 2, H-1117, HUNGARY
szirmay@fsz.bme.hu
Abstract:
Fluid dynamics is described by the Navier-Stokes differential equations, which need to be solved by fluid simulators. The numerical solution of these equations in 3D has high computational and data storing costs. In this paper we present a real-time fluid simulation and visualization method that exploits the computational power of a GPU cluster to solve this task interactively. We use object space decomposition of the 3D volume. A single CPU/GPU node solves the Navier-Stokes equations only for its subvolume, taking into account the data from neighboring nodes as boundary conditions. Mathematically, the numerical simulation is executed on a regular grid, i.e. an Eulerian approach is taken. Fluid data at the grid points, including the velocity field and a custom display variable are stored in specially organized 3D textures. In order to guarantee the stability of the solution we have adopted the Stable Fluid approach in our distributed solution, which have been extended by vorticity confinement and level of detail control using the early z-culling feature of the graphics hardware. The GPU of a node executes both the simulation and the rendering of the subvolume. We used a texture slicing rendering method with alpha blending to display a custom scalar field that is carried around by the simulated flow in the subvolume. The images of different nodes are composited together by the parallel pipeline algorithm of HP's ParaComp library. Our implementation has good scalability and makes the simulation of larger data sets possible at acceptable frame rates.
Keywords:
Fluid dynamics, GPU simulation, GPU cluster, ParaComp, HP-SVA.