high performance computing on graphics processing units: hgpu.org

Navier-Stokes Equations are a mathematical model to describe the behaviour of fluids. They have proven to represent real fluid flows quite well and are base for many fluid simulations. In order to exploit the performance provided by modern many-core systems, fluid simulation algorithms must be able to efficiently solve the Navier-Stokes Equations in parallel. The aim of this interdisciplinary project was to implement a real-time fluid simulation by solving the Navier-Stokes Equations on a GPU. Additionally, a user should have the possibility to interact with the simulation at run-time, requiring real-time visualization. The sequential algorithm described by Michael Griebel [1] provides the basis for the implemented simulation. Within the project a solver for CPUs was implemented and ported to GPUs. It is capable of simulating a two-dimensional laminar flow and visualize it in real time. Obstacles can be added at run-time by drawing with the cursor on the visualization window and have immediate effects on the simulation. The simulation framework and a CPU solver was written in C++, while OpenCL was used for kernels of the GPU solver. The user interface and visualization are done with the help of the Qt libraries and OpenGL.