Abstract
In recent years, there has been a drive towards parallel architectures to further
increase computational performance. The many-core architecture of
modern Graphics Processing Units (GPU) can be used for general computations
in addition to graphics processing, and provide good performance for
data parallel computations.
In this thesis, we explore the processing powers of two generations of GPUs
by implementing H.264 prediction algorithms. We have implemented motion
vector search and motion vector prediction on the GPU, and discuss how
they to the parallel architecture.