Efficiency of Parallelization of Neural Network Algorithm on Graphic Cards