Numerical solution of fluid mechanics problems on general-purpose graphics processor units

Possibilities of using general-purpose graphics processor units for the solution of fluid mechanics problems are discussed. The CUDA technology is used to implement parallel computational algorithms. A number of model problems are solved on graphics processor. Some approaches to the optimization of programming codes related to the use of various types of memory are considered. Features of implementation of the splitting scheme (projection method) used to simulate the flows of viscous incompressible fluid are analyzed. The speed-up of solution on graphics processor units is compared with the solution on central processor units with consideration of different meshes and different methods of partitioning of input data into blocks.