I have the impression that at least the training phase of alphago needs much resources to achieve superior results, though it would be nice to see some "exact"

I have the impression that at least the training phase of alphago needs much resources to achieve superior results, though it would be nice to see some "exact" numbers. IMO combined limit for runtime+training energy would be the best way to keep things interesting.