N.G. Andrianov

Processing of images is a labor-consuming computing problem and not desktop computer is required powerful computing technicians for its decision in real time. Making some operations in parallel for all pixels of a shot can be reduced considerably time. Using special calculators and supercomputers expensively, take a lot of place and demand programs specially written under them. Recently had the big development graphic processors which will be used for processing of images. On modern video cards the multi core processors meant for performance of various graphic problems with use of thousand of streams are established. The number of executing kernels begins from 96 and reaches 480. The possibilities offered by graphic processors which actually are powerful SIMD (Single Instruction Multiple Data) processors are considered. For work with such processors recently there was new technology CUDA (Compute Unified Device Architecture) which allows writing usual programs on expanded with. The thought over architecture of the processor allows is high-grade to use all its possibilities. There is the comparative analysis of work of two algorithms and processing of the image, each of which is realized for work on CPU and GPU is carried out in work. The basic possibilities of graphic processors are studied. The results show the big superiority of graphic processors in front of usual processors for the of difficult computing problems' decision.