Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems

Karlo Kraljic; Daniel Kerger; Martin Schulz

doi:10.1007/978-3-031-21867-5_1

Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems

Karlo Kraljic
, Daniel Kerger
, Martin Schulz

Informatics 10 - Chair of Computer Architecture and Parallel Systems

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

12 Scopus citations

Abstract

With most major corporations and research institutions having pledged to support sustainability goals for High Performance Computing (HPC), energy efficiency is a critical factor when evaluating heterogeneous HPC systems. However, many popular hardware performance & energy measurement frameworks, such as LIKWID, and benchmarks, such as the STREAM or the hipBone benchmark, do not or not fully support execution on heterogeneous systems containing AMD or NVIDIA Graphical Processing Units (GPUs), leading to a gap with regards to the understanding the relationship between frequency, performance and energy. We aim at closing this gap by extending the performance measurement framework LIKWID to support both AMD and NVIDIA GPUs. We run the STREAM and hipBone benchmark on AMD and NVIDIA GPUs at different GPU core frequencies. We show that the minimum period between two measurements for our GPU is at least 100ms and that GPUs have a sweet spot with regards to energy consumption at approximately 75% of their maximum frequency with energy savings up to 30% at a performance overhead between 0.72% and 3.12%.

Original language	English
Title of host publication	Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings
Editors	Martin Schulz, Carsten Trinitis, Nikela Papadopoulou, Thilo Pionteck
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	3-16
Number of pages	14
ISBN (Print)	9783031218668
DOIs	https://doi.org/10.1007/978-3-031-21867-5_1
State	Published - 2022
Event	35th International Conference on Architecture of Computing Systems, ARCS 2022 - Heilbronn, Germany Duration: 13 Sep 2022 → 15 Sep 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13642 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	35th International Conference on Architecture of Computing Systems, ARCS 2022
Country/Territory	Germany
City	Heilbronn
Period	13/09/22 → 15/09/22

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

Energy awareness
Heterogeneous computing
High Performance Computing

Access to Document

10.1007/978-3-031-21867-5_1

Cite this

Kraljic, K., Kerger, D., & Schulz, M. (2022). Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems. In M. Schulz, C. Trinitis, N. Papadopoulou, & T. Pionteck (Eds.), Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings (pp. 3-16). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13642 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-21867-5_1

Kraljic, Karlo ; Kerger, Daniel ; Schulz, Martin. / Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems. Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. editor / Martin Schulz ; Carsten Trinitis ; Nikela Papadopoulou ; Thilo Pionteck. Springer Science and Business Media Deutschland GmbH, 2022. pp. 3-16 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{10fce7c91c3a4952b7ccf00bcf51193a,

title = "Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems",

abstract = "With most major corporations and research institutions having pledged to support sustainability goals for High Performance Computing (HPC), energy efficiency is a critical factor when evaluating heterogeneous HPC systems. However, many popular hardware performance \& energy measurement frameworks, such as LIKWID, and benchmarks, such as the STREAM or the hipBone benchmark, do not or not fully support execution on heterogeneous systems containing AMD or NVIDIA Graphical Processing Units (GPUs), leading to a gap with regards to the understanding the relationship between frequency, performance and energy. We aim at closing this gap by extending the performance measurement framework LIKWID to support both AMD and NVIDIA GPUs. We run the STREAM and hipBone benchmark on AMD and NVIDIA GPUs at different GPU core frequencies. We show that the minimum period between two measurements for our GPU is at least 100ms and that GPUs have a sweet spot with regards to energy consumption at approximately 75\% of their maximum frequency with energy savings up to 30\% at a performance overhead between 0.72\% and 3.12\%.",

keywords = "Energy awareness, Heterogeneous computing, High Performance Computing",

author = "Karlo Kraljic and Daniel Kerger and Martin Schulz",

note = "Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 35th International Conference on Architecture of Computing Systems, ARCS 2022 ; Conference date: 13-09-2022 Through 15-09-2022",

year = "2022",

doi = "10.1007/978-3-031-21867-5\_1",

language = "English",

isbn = "9783031218668",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "3--16",

editor = "Martin Schulz and Carsten Trinitis and Nikela Papadopoulou and Thilo Pionteck",

booktitle = "Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings",

}

Kraljic, K, Kerger, D & Schulz, M 2022, Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems. in M Schulz, C Trinitis, N Papadopoulou & T Pionteck (eds), Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13642 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 3-16, 35th International Conference on Architecture of Computing Systems, ARCS 2022, Heilbronn, Germany, 13/09/22. https://doi.org/10.1007/978-3-031-21867-5_1

Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems. / Kraljic, Karlo; Kerger, Daniel; Schulz, Martin.
Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. ed. / Martin Schulz; Carsten Trinitis; Nikela Papadopoulou; Thilo Pionteck. Springer Science and Business Media Deutschland GmbH, 2022. p. 3-16 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13642 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems

AU - Kraljic, Karlo

AU - Kerger, Daniel

AU - Schulz, Martin

PY - 2022

Y1 - 2022

N2 - With most major corporations and research institutions having pledged to support sustainability goals for High Performance Computing (HPC), energy efficiency is a critical factor when evaluating heterogeneous HPC systems. However, many popular hardware performance & energy measurement frameworks, such as LIKWID, and benchmarks, such as the STREAM or the hipBone benchmark, do not or not fully support execution on heterogeneous systems containing AMD or NVIDIA Graphical Processing Units (GPUs), leading to a gap with regards to the understanding the relationship between frequency, performance and energy. We aim at closing this gap by extending the performance measurement framework LIKWID to support both AMD and NVIDIA GPUs. We run the STREAM and hipBone benchmark on AMD and NVIDIA GPUs at different GPU core frequencies. We show that the minimum period between two measurements for our GPU is at least 100ms and that GPUs have a sweet spot with regards to energy consumption at approximately 75% of their maximum frequency with energy savings up to 30% at a performance overhead between 0.72% and 3.12%.

AB - With most major corporations and research institutions having pledged to support sustainability goals for High Performance Computing (HPC), energy efficiency is a critical factor when evaluating heterogeneous HPC systems. However, many popular hardware performance & energy measurement frameworks, such as LIKWID, and benchmarks, such as the STREAM or the hipBone benchmark, do not or not fully support execution on heterogeneous systems containing AMD or NVIDIA Graphical Processing Units (GPUs), leading to a gap with regards to the understanding the relationship between frequency, performance and energy. We aim at closing this gap by extending the performance measurement framework LIKWID to support both AMD and NVIDIA GPUs. We run the STREAM and hipBone benchmark on AMD and NVIDIA GPUs at different GPU core frequencies. We show that the minimum period between two measurements for our GPU is at least 100ms and that GPUs have a sweet spot with regards to energy consumption at approximately 75% of their maximum frequency with energy savings up to 30% at a performance overhead between 0.72% and 3.12%.

KW - Energy awareness

KW - Heterogeneous computing

KW - High Performance Computing

UR - https://www.scopus.com/pages/publications/85144814451

U2 - 10.1007/978-3-031-21867-5_1

DO - 10.1007/978-3-031-21867-5_1

M3 - Conference contribution

AN - SCOPUS:85144814451

SN - 9783031218668

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 3

EP - 16

BT - Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings

A2 - Schulz, Martin

A2 - Trinitis, Carsten

A2 - Papadopoulou, Nikela

A2 - Pionteck, Thilo

PB - Springer Science and Business Media Deutschland GmbH

T2 - 35th International Conference on Architecture of Computing Systems, ARCS 2022

Y2 - 13 September 2022 through 15 September 2022

ER -

Kraljic K, Kerger D, Schulz M. Energy Efficient Frequency Scaling on GPUs in Heterogeneous HPC Systems. In Schulz M, Trinitis C, Papadopoulou N, Pionteck T, editors, Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. Springer Science and Business Media Deutschland GmbH. 2022. p. 3-16. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-21867-5_1