PARALLELIZATION OF NEURAL NETWORK ALGORITHM FOR BOUNDARY PROBLEMS SOLUTION                        IN MATHEMATICAL PHYSICS ON MULTI-CORE PROCESSORS NVIDIA 

Gorbachenko Vladimir Ivanovich, Doctor of engineering sciences, professor, head of sub-department of informatics and computing systems, Penza State Pedagogical University named after V. G. Belinsky, gorvi@mail.ru
Matveeva Natalya Olegovna, Programmer, automation department, Penza State Pedagogical University named after V. G. Belinsky, gorvi@mail.ru
Gurin Evgeny Ivanovich, Doctor of engineering sciences, professor, sub-department of computer engineering, Penza State University, gurin2@yandex.ru 

004.272:004.032.26 

The article describes the possibility of effective realization of the algorithm for solving boundary value problems of mathematical physics on multicore Graphics Processing Unit (GPU) NVIDIA GeForce 8800 GTX. The authors propose parallel realization of the training algorithm of the radial basis neural network (RBFNN), using CUDA. The proceeding compares execution time using CPU versus GPU. 

massive parallel processing, Graphics Processing Unit, parallelism, CUDA, radial basis neural network, differential equation.