There may be some utility to optimizing a given neural network to exploit the features of a given computational architecture. The point would be to expand the

There may be some utility to optimizing a given neural network to exploit the features of a given computational architecture. The point would be to expand the width of each layer to hit the limits of your hardware's parallelism. I.e., partition the network into n parts and shallow-ize each one to produce a new neural network with n layers, each of which being roughly as wide as your p parallel computational channels.