The emergence of stream applications such as multimedia processing and networking made Stream processing to become increasingly important. In a stream programming model, computations are decomposed into stages of sequential computations and input/output i partitioned into chunks where each chunk becomes a stream element. Input streams are then passed through stages of a pipeline, producing output streams. This has the advantage of effectively using the inherent parallelism in the pipeline, i.e., the execution of the stages can be overlapped with each other to exploit task parallelism.

The objective of the thesis is to investigate the suitability of Graphics Processing Unit(GPU) for stream parallel applications and to find out an optimal number of streams so that the computation on a chunk fully utilizes the GPU resources and the completion time of the whole application is the minimum one that can be achieved due to the usage of CUDA streams.