Graphics Processing Units have received much attention in the last years. Using explicit algorithms that exploit the computational power of these devices, it was possible to develop fluid solvers that are much faster than their CPU counterparts. However these explicit solvers usually rely on artificially slow sound speed, which in many cases translates into the loss of the physics of the problem.
In the RealTime project, a fractional-step scheme (splitting the velocity and pressure into two different equations) was used to overcome this problem and allow for incompressible fluids. The most important tool developed in the project was an FFT preconditioner combined with a CG solver. This allows to solved efficienctly the Poisson pressure equations with algorithms that make use of the full potential of the GPU, which translates into extremely fast solvers that yet remain accurate. Unfortunately, the developed strategy is only valid for structured, cartesian meshes.
Below some examples are presented.