USING NEURAL NETWORKS IN PARALLEL SYSTEMS AND GPU

The article examines the results of ten years of work in the field of parallelizing the operation of neural networks in parallel and vector computer architectures. The research was conducted on the cluster of the G.E. Pukhov Institute of Modeling Problems in Energy Engineering of the National Academy of Sciences of Ukraine (Intel Xeon 5405, 4×2 GB DDR-2, InfiniBand 20 Gbps), the cluster of Zaporizhzhia National Technical University (Intel E3200, 1 GB DDR-2, Gigabit Ethernet 1 Gbps), as well as on NVIDIA GTX 285+ (240 CUDA cores) and NVIDIA GTX 960 (1024 CUDA cores) graphics accelerators. The article provides data on the runtime of neural networks on a single core and multiple cores, illustrating the acceleration of the computational process. A conclusion is drawn regarding the feasibility of parallelizing neural network operations. Keywords: neuro-fuzzy network, parallel computing, diagnostics, forecasting, genetic algorithm, production rules, multi-agent method.