Fine-Grained Power Management for

This paper proposes a new hardware-based power management technique for multithreaded processors that schedule instructions by the Guaranteed Percentage real-time scheduling scheme, which assigns percentages of the processors utilization to the di#erent threads. Our power management technique uses the sum of the percentages of the currently active threads to control frequency reduction, dynamic voltage scaling, and pipeline gating. Results are obtained by simulating the execution of four threads derived from an autonomous guided vehicle application in a specific processor core -- the Komodo processor. Our evaluation showed for that benchmark an average processor utilization of 22.6% and a frequent change of utilization in the range of 0% to 58%; energy consumption could be reduced to 12.7% of the energy required by a system running at top speed