GPU schedulers: how fair is fair enough?