Errors-in-variables identification of thermal models for many-core computing systems

Roberto Diversi, Andrea Tilli, Andrea Bartolini, Luca Benini

Abstract: A crucial issue in next generation many-core computing systems is the dynamic thermal control for run-time performance optimization. To this end, scalable, compact and effective thermal models are essential. In this paper, we present an algorithm based on the Frisch scheme to obtain identification of distributed core-centric interacting models, dealing with very noisy temperature measurements and high process noises. We apply the proposed solution to an Intel's Single-chip-Cloud-Computer (SCC), a many-core prototype with 48 cores.