Optimizing energy consumption in GPUS through feedback-driven CTA scheduling

Amin Jadidi, Mohammad Arjomand, Mahmut Kandemir, Chitaranjan Das

Research output: Contribution to journalConference article

Abstract

Emerging GPU architectures offer a cost-effective computing platform by providing thousands of energy-efficient compute cores and high bandwidth memory that facilitate the execution of highly parallel applications. In this paper, we show that different applications, and in fact different kernels from the same application might exhibit significantly varying utilizations of compute and memory resources. In order to improve the energy efficiency of the GPU system, we propose a run-time characterization strategy that classifies kernels as compute- or memory-intensive based on their resource utilizations. Using this knowledge, our proposed mechanism employs core shut-down technique for memory-intensive kernels in order to manage energy in a more efficient way. This strategy uses performance and memory bandwidth utilization information to determine the ideal hardware configuration at run-time. The proposed technique saves on average 21% of total chip energy for memory-intensive applications, which is within 8% of the optimal saving that can be obtained from an oracle scheme.

Original languageEnglish (US)
Pages (from-to)129-140
Number of pages12
JournalSimulation Series
Volume49
Issue number3
StatePublished - Jan 1 2017
Event25th High Performance Computing Symposium, HPC 2017, Part of the 2017 Spring Simulation Multi-Conference, SpringSim 2017 - Virginia Beach, United States
Duration: Apr 23 2017Apr 26 2017

Fingerprint

Energy utilization
Scheduling
Feedback
Data storage equipment
Bandwidth
Computer hardware
Energy efficiency
Costs
Graphics processing unit

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications

Cite this

@article{861ab6f8c41b435e9651a79a6c397762,
title = "Optimizing energy consumption in GPUS through feedback-driven CTA scheduling",
abstract = "Emerging GPU architectures offer a cost-effective computing platform by providing thousands of energy-efficient compute cores and high bandwidth memory that facilitate the execution of highly parallel applications. In this paper, we show that different applications, and in fact different kernels from the same application might exhibit significantly varying utilizations of compute and memory resources. In order to improve the energy efficiency of the GPU system, we propose a run-time characterization strategy that classifies kernels as compute- or memory-intensive based on their resource utilizations. Using this knowledge, our proposed mechanism employs core shut-down technique for memory-intensive kernels in order to manage energy in a more efficient way. This strategy uses performance and memory bandwidth utilization information to determine the ideal hardware configuration at run-time. The proposed technique saves on average 21{\%} of total chip energy for memory-intensive applications, which is within 8{\%} of the optimal saving that can be obtained from an oracle scheme.",
author = "Amin Jadidi and Mohammad Arjomand and Mahmut Kandemir and Chitaranjan Das",
year = "2017",
month = "1",
day = "1",
language = "English (US)",
volume = "49",
pages = "129--140",
journal = "Simulation Series",
issn = "0735-9276",
number = "3",

}

Optimizing energy consumption in GPUS through feedback-driven CTA scheduling. / Jadidi, Amin; Arjomand, Mohammad; Kandemir, Mahmut; Das, Chitaranjan.

In: Simulation Series, Vol. 49, No. 3, 01.01.2017, p. 129-140.

Research output: Contribution to journalConference article

TY - JOUR

T1 - Optimizing energy consumption in GPUS through feedback-driven CTA scheduling

AU - Jadidi, Amin

AU - Arjomand, Mohammad

AU - Kandemir, Mahmut

AU - Das, Chitaranjan

PY - 2017/1/1

Y1 - 2017/1/1

N2 - Emerging GPU architectures offer a cost-effective computing platform by providing thousands of energy-efficient compute cores and high bandwidth memory that facilitate the execution of highly parallel applications. In this paper, we show that different applications, and in fact different kernels from the same application might exhibit significantly varying utilizations of compute and memory resources. In order to improve the energy efficiency of the GPU system, we propose a run-time characterization strategy that classifies kernels as compute- or memory-intensive based on their resource utilizations. Using this knowledge, our proposed mechanism employs core shut-down technique for memory-intensive kernels in order to manage energy in a more efficient way. This strategy uses performance and memory bandwidth utilization information to determine the ideal hardware configuration at run-time. The proposed technique saves on average 21% of total chip energy for memory-intensive applications, which is within 8% of the optimal saving that can be obtained from an oracle scheme.

AB - Emerging GPU architectures offer a cost-effective computing platform by providing thousands of energy-efficient compute cores and high bandwidth memory that facilitate the execution of highly parallel applications. In this paper, we show that different applications, and in fact different kernels from the same application might exhibit significantly varying utilizations of compute and memory resources. In order to improve the energy efficiency of the GPU system, we propose a run-time characterization strategy that classifies kernels as compute- or memory-intensive based on their resource utilizations. Using this knowledge, our proposed mechanism employs core shut-down technique for memory-intensive kernels in order to manage energy in a more efficient way. This strategy uses performance and memory bandwidth utilization information to determine the ideal hardware configuration at run-time. The proposed technique saves on average 21% of total chip energy for memory-intensive applications, which is within 8% of the optimal saving that can be obtained from an oracle scheme.

UR - http://www.scopus.com/inward/record.url?scp=85020624045&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85020624045&partnerID=8YFLogxK

M3 - Conference article

VL - 49

SP - 129

EP - 140

JO - Simulation Series

JF - Simulation Series

SN - 0735-9276

IS - 3

ER -