It does seem rather underpowered if you compare it with the requirements of state of the art Deep Learning models in other areas, not an apples and oranges comparison though.