No description available.

Fluids such as smoke, water and fire are simulated for both Computer Graphics applications and engineering fields such as Mechanical Engineering. Generally, Fluid Dynamics is used for the achievement of realistic-looking fluid simulations. However, the complexity of these calculations makes it difficult to achieve high performance. With the advances in graphics hardware, it has been possible to provide programmability both at the vertex and the fragment level, which allows for faster simulations of complex fluids and other events. In this thesis, one gaseous fluid, smoke is simulated in three dimensions by solving Navier-Stokes Equations (NSEs) using a semi-Lagrangian unconditionally stable method. Simulation is performed both on Central Processing Unit (CPU) and Graphics Processing Unit (GPU). For the programmability at the vertex and the fragment level, C for Graphics (Cg), a platform-independent and architecture neutralshading language, is used. Owing to the advantage of programmability and parallelism of GPU, smoke simulation on graphics hardware runs significantly faster than the corresponding CPU implementation. The test results prove the higher performance of GPU over CPU for running three dimensional fluid simulations.