Due to a constant increase of need for computational power, additional approaches to classical processors has been developed. Computational workload is more and more solved by using massive parallel computer architectures. This paper contains a short historical overview of development and features of such computers. Described are theoretical facts and boundaries.
Second part consists of a practical example of computer vision application for industrial purposes and speed measurements of specific functions on an embedded platform. Algorithm is written using Nvidia CUDA implementation of OpenCV functions. This allows us to employ all cores of an integrated graphic processor.