There’s also the CPU. Those <=200 years of SC2 simulations per agent aren’t free. OA5, recall, was ’256 GPUs and 128,000 CPU cores’. (Occasionally training a small NN update is easier than running many games necessary to get the experience to decide what tweak to make.)
There’s also the CPU. Those <=200 years of SC2 simulations per agent aren’t free. OA5, recall, was ’256 GPUs and 128,000 CPU cores’. (Occasionally training a small NN update is easier than running many games necessary to get the experience to decide what tweak to make.)