Self-Aware Thermal Management for High-Performance Computing Processors

Editor’s note:
Thermal management in high-performance multicore platforms has become exceedingly complex due to variable workloads, thermal heterogeneity, and long, thermal transients. This article addresses these complexities by sophisticated analysis of noisy thermal sensor readings, dynamic learning to adapt to the peculiarities of the hardware and the applications, and a dynamic optimization strategy.
—Axel Jantsch, TU Wien
—
Nikil Dutt, University of California at Irvine