If more compute is spent than before but in parallel to reduce latency then wouldn't this increase power? Shouldn't latency and power both be objective?