Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling

Da Cheng Juan, Siddharth Garg, Jinpyo Park, Diana Marculescu

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Near-Threshold Computing (NTC) has emerged as a solution that promises to significantly increase the energy efficiency of next-generation multi-core systems. This paper evaluates and analyzes the behavior of dynamic voltage and frequency scaling (DVFS) control algorithms for multi-core systems operating under near-threshold, nominal, or turbo-mode conditions. We adapt the model selection technique from machine learning to learn the relationship between performance and power. The theoretical results show that the resulting models satisfy convexity properties essential to efficiently determining optimal voltage/frequency operating points for minimizing energy consumption under throughput constraints or maximizing throughput under a given power budget. Our experimental results show that, compared with DVFS in the conventional operating range, extended range DVFS control including turbo-mode and near-threshold operation achieves an additional (1) 13.28% average energy reduction under isoperformance conditions, and (2) 7.54% average throughput increase under iso-power conditions.

Original languageEnglish (US)
Title of host publication2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013
PublisherIEEE Computer Society
Pages1-10
Number of pages10
ISBN (Print)9781479914173
DOIs
StatePublished - 2013
Event11th ACM/IEEE International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013 - Montreal, QC, Canada
Duration: Sep 29 2013Oct 4 2013

Other

Other11th ACM/IEEE International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013
CountryCanada
CityMontreal, QC
Period9/29/1310/4/13

Fingerprint

Throughput
Electric potential
Energy efficiency
Learning systems
Energy utilization
Dynamic frequency scaling
Voltage scaling

Keywords

  • Chip-multiprocessor
  • Convex optimization
  • Dynamic voltage and frequency scaling
  • Machine learning
  • Power management

ASJC Scopus subject areas

  • Software
  • Hardware and Architecture

Cite this

Juan, D. C., Garg, S., Park, J., & Marculescu, D. (2013). Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling. In 2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013 (pp. 1-10). [6658995] IEEE Computer Society. https://doi.org/10.1109/CODES-ISSS.2013.6658995

Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling. / Juan, Da Cheng; Garg, Siddharth; Park, Jinpyo; Marculescu, Diana.

2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013. IEEE Computer Society, 2013. p. 1-10 6658995.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Juan, DC, Garg, S, Park, J & Marculescu, D 2013, Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling. in 2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013., 6658995, IEEE Computer Society, pp. 1-10, 11th ACM/IEEE International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013, Montreal, QC, Canada, 9/29/13. https://doi.org/10.1109/CODES-ISSS.2013.6658995
Juan DC, Garg S, Park J, Marculescu D. Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling. In 2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013. IEEE Computer Society. 2013. p. 1-10. 6658995 https://doi.org/10.1109/CODES-ISSS.2013.6658995
Juan, Da Cheng ; Garg, Siddharth ; Park, Jinpyo ; Marculescu, Diana. / Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling. 2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013. IEEE Computer Society, 2013. pp. 1-10
@inproceedings{95bba4c7e14d4706bcc20d14cea02dc1,
title = "Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling",
abstract = "Near-Threshold Computing (NTC) has emerged as a solution that promises to significantly increase the energy efficiency of next-generation multi-core systems. This paper evaluates and analyzes the behavior of dynamic voltage and frequency scaling (DVFS) control algorithms for multi-core systems operating under near-threshold, nominal, or turbo-mode conditions. We adapt the model selection technique from machine learning to learn the relationship between performance and power. The theoretical results show that the resulting models satisfy convexity properties essential to efficiently determining optimal voltage/frequency operating points for minimizing energy consumption under throughput constraints or maximizing throughput under a given power budget. Our experimental results show that, compared with DVFS in the conventional operating range, extended range DVFS control including turbo-mode and near-threshold operation achieves an additional (1) 13.28{\%} average energy reduction under isoperformance conditions, and (2) 7.54{\%} average throughput increase under iso-power conditions.",
keywords = "Chip-multiprocessor, Convex optimization, Dynamic voltage and frequency scaling, Machine learning, Power management",
author = "Juan, {Da Cheng} and Siddharth Garg and Jinpyo Park and Diana Marculescu",
year = "2013",
doi = "10.1109/CODES-ISSS.2013.6658995",
language = "English (US)",
isbn = "9781479914173",
pages = "1--10",
booktitle = "2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013",
publisher = "IEEE Computer Society",

}

TY - GEN

T1 - Learning the optimal operating point for many-core systems with extended range voltage/frequency scaling

AU - Juan, Da Cheng

AU - Garg, Siddharth

AU - Park, Jinpyo

AU - Marculescu, Diana

PY - 2013

Y1 - 2013

N2 - Near-Threshold Computing (NTC) has emerged as a solution that promises to significantly increase the energy efficiency of next-generation multi-core systems. This paper evaluates and analyzes the behavior of dynamic voltage and frequency scaling (DVFS) control algorithms for multi-core systems operating under near-threshold, nominal, or turbo-mode conditions. We adapt the model selection technique from machine learning to learn the relationship between performance and power. The theoretical results show that the resulting models satisfy convexity properties essential to efficiently determining optimal voltage/frequency operating points for minimizing energy consumption under throughput constraints or maximizing throughput under a given power budget. Our experimental results show that, compared with DVFS in the conventional operating range, extended range DVFS control including turbo-mode and near-threshold operation achieves an additional (1) 13.28% average energy reduction under isoperformance conditions, and (2) 7.54% average throughput increase under iso-power conditions.

AB - Near-Threshold Computing (NTC) has emerged as a solution that promises to significantly increase the energy efficiency of next-generation multi-core systems. This paper evaluates and analyzes the behavior of dynamic voltage and frequency scaling (DVFS) control algorithms for multi-core systems operating under near-threshold, nominal, or turbo-mode conditions. We adapt the model selection technique from machine learning to learn the relationship between performance and power. The theoretical results show that the resulting models satisfy convexity properties essential to efficiently determining optimal voltage/frequency operating points for minimizing energy consumption under throughput constraints or maximizing throughput under a given power budget. Our experimental results show that, compared with DVFS in the conventional operating range, extended range DVFS control including turbo-mode and near-threshold operation achieves an additional (1) 13.28% average energy reduction under isoperformance conditions, and (2) 7.54% average throughput increase under iso-power conditions.

KW - Chip-multiprocessor

KW - Convex optimization

KW - Dynamic voltage and frequency scaling

KW - Machine learning

KW - Power management

UR - http://www.scopus.com/inward/record.url?scp=84892633011&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84892633011&partnerID=8YFLogxK

U2 - 10.1109/CODES-ISSS.2013.6658995

DO - 10.1109/CODES-ISSS.2013.6658995

M3 - Conference contribution

SN - 9781479914173

SP - 1

EP - 10

BT - 2013 International Conference on Hardware/Software Codesign and System Synthesis, CODES+ISSS 2013

PB - IEEE Computer Society

ER -