Evaluating Performance, Power and Energy of Deep Neural Networks on CPUs and GPUs