Abstract: Described is a process for optimizing energy utilization in a computer processing device, including sampling energy and power measurements for given hardware components, storing the energy and power measurements, computing metrics for a sample based on current power consumption and current power limits, comparing the current sample metrics against a metric threshold, classifying the current sample based on the comparison, assigning a classification type based on the classifying the current sample, determining if an actual number of a computational intensity characteristic exceeds a maximum allowed reference number in a sample window for the classification type, if the actual number of a computational intensity characteristic exceeds the maximum for the classification type, computing a new maximum allowed for the classification type, and constraining the hardware components to the new maximum energy and power measurements.

Abstract: Described is a process for optimizing energy utilization in a computer processing device, including sampling energy and power measurements for given hardware components, storing the energy and power measurements, computing metrics for a sample based on current power consumption and current power limits, comparing the current sample metrics against a metric threshold, classifying the current sample based on the comparison, assigning a classification type based on the classifying the current sample, determining if an actual number of a computational intensity characteristic exceeds a maximum allowed reference number in a sample window for the classification type, if the actual number of a computational intensity characteristic exceeds the maximum for the classification type, computing a new maximum allowed for the classification type, and constraining the hardware components to the new maximum energy and power measurements.