Thanks for the great article. I’ve got a question about RAM. I haven't had a chance to run GPUs based simulations, but from my experience and types of ML simulations I were running on CPU, I required 1GB per core. What would you say about the requirements fro GPUs ? I know it depends way too much on the data(is it numerical, multimedia, how much of it etc) as well as what the computational graph is itself, but I would still be very interested to hear about your thinking on this.