SmartDPM: Machine learning-based dynamic power management for multi-core microprocessors

P. D.Sai Manoj, Axel Jantsch, Muhammad Shafique

Research output: Contribution to journalArticlepeer-review

Abstract

To address the power management challenge in multi-core microprocessors, we present a lightweight machine learning based dynamic power management (SmartDPM) scheme in which the voltage-frequency levels of the cores are dynamically adjusted along with online learning based workload prediction in an observer-controller loop. To enable scalability, our SmartDPM employs a per-application autonomous power management policy, in which online machine learning principles are employed for predicting the workload and capturing sporadic variations under the constraints of accurate yet lightweight. Further, applications are assigned appropriate voltage-frequency level towards an efficient power management. The learning helps in dynamically reducing prediction error. Compared to the non-DVFS implementation, SmartDPM achieves nearly 35% power saving and nearly 15% higher power savings on average compared to the existing machine learning based power management schemes for a microprocessor with up to 32-cores.

Original languageEnglish (US)
Pages (from-to)460-474
Number of pages15
JournalJournal of Low Power Electronics
Volume14
Issue number4
DOIs
StatePublished - Dec 2018

Keywords

  • Control theory
  • Dynamic voltage frequency scaling
  • Energy efficiency
  • Machine learning
  • Multi-core microprocessor
  • Online learning
  • Power management
  • Prediction
  • Scalability

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Fingerprint Dive into the research topics of 'SmartDPM: Machine learning-based dynamic power management for multi-core microprocessors'. Together they form a unique fingerprint.

Cite this