I feel like this ignores the complexity of the distributed training frameworks. The challenge is in making it fast at scale.