Optimize CPU-GPU Power Allocation with Parameter and Action Networks

The power allocation mechanism between the CPU and GPU significantly impacts the performance of laptops. This paper proposes a novel method for optimizing CPU-GPU power allocation. Specifically, it defines two networks: one, termed ActionNet, is responsible for controlling power allocation; the other, named ParameterNet, is used to select the optimal parameters of ActionNet. This paper proposes a process involving sampling, evaluation, training, and selection to identify the best parameters for ActionNet. Experimental results demonstrate that this method can find a suitable set of parameters for ActionNet, effectively controlling CPU-GPU power allocation and enhancing system performance without exceeding temperature limits.