A High Energy Efficiency Algorithm For Microprocessor Based On Real-Time Power Budget

With the reduction of the characteristic size of integrated circuits,higher integration density brings higher power density.Due to the limitation of chip heat dissipation,multi-core and multi-core microprocessors have finally entered the era of dark silicon.There are two outstanding problems in the dark silicon era,one is how to maximize the performance at the safe threshold temperature,and the other is how to optimize the energy efficiency ratio of multi-core chips.The difficulty of energy efficiency optimization of multi-core chips is that the proportion of static power in the total power increases gradually with the decrease of IC feature size,so it is not feasible to ignore the static power or consider the static power as a constant in the past.Moreover,due to the thermal sensitivity of static power consumption,it is inevitable to introduce a thermal model when solving such a problem.The temperature of any given point of a multi-core chip is affected by the work-load distribution and the complexity of the task,so it is very complex or even impossible to solve such a problem directly.On the basis of the above,this paper proposes a multi-core microprocessor energy ef-ficiency optimization algorithm,which considers the nonlinear relationship between tem-perature and power accurately.This algorithm mainly establishes an accurate static power model,and in order to solve the temperature coupling effect of multi-core chip,it also es-tablishes thermal model,performance model and other models,and combines the involved parameters to form a system Finally,it is solved by gradient descent algorithm.In order to prove the applicability and correctness of the proposed algorithm,this paper also uses the Monte Carlo scanning algorithm to calculate the position of the energy efficiency ratio in all possible distributions.Finally,the experimental results show that the proposed algorithm can accurately solve the problem of energy efficiency ratio optimiza-tion of multi-core microprocessor within acceptable time.Since there is a corresponding relationship between frequency and power,and the thermal management related simula-tion in this paper usually takes power as the management condition,the state of energy efficiency ratio can also be adjusted by calculating the power budget.