Massively Parallel Reinforcement Learning With An Application To Video Games