Not every computing platform is expected to do fast inference.