When Will General Purpose Micro-processors Simulate Neural Networks in Real Time for Hep Applications ?

By their universal character, general purpose microprocessors may be used to simulate arti cial neural networks. However, until now, they were not capable to perform these simulations in real-time. On the other hand, the computational power of these processors has tremendously increased recently. Thus, one may wonder whether up-to-date general purpose micro processors can simulate neural networks in real-time. To answer this question, we need to evaluate the performances of these architectures for the simulation of neural networks. To realize this evaluation we have developed an original methodology [7] which can predict the simulation time of a neural network on an electronic architecture. This prediction is based on an analytic model of the architecture performances.