The CNN B has fewer convolutional layers than the Starting CNN. The fully-connected layer is exactly the same in both networks. Hence, the two CNNs have similar numbers of weights but the CNN B is about 11–13 times faster (depending on CPU/GPU processing). The absolute running time values show that the time gain becomes really essential when the processing is done by a CPU, which is often the case in big computational clusters or when images are processed separately rather than in batches (in the latter case, an impact from GPUs is negligible).