Perhaps the incentive is for variable behavior. When there is low GPU demand, burn more, but reduce when there is contention.
Perhaps the incentive is for variable behavior. When there is low GPU demand, burn more, but reduce when there is contention.