The high-speed method of calculation of depth card of the stereoimage which each pixel contains the information on distance up to a stereocamera, for realization on graphic processor (GPU) is offered. The given decision is based on use of several sliding windows of adaptive structure with pre-and postprocessing of entrance and target images. High efficiency is achieved due to architecture modern GP, the most suitable for the decision of the problem of depth card calculation by local methods. The carried comparison of an operating time of algorithm on central processor (CP) and on GP shows high scalability of the suggested decision. High speed will allow to use the given method in problems of real time, for example, in a robotics and creation of natural interfaces. Ways of probable improvement of the given method are discussed.