Multi-Prediction Compression: An Efficient and Scalable Memory Compression Framework for GP-GPU

Hoyong Jin; Donghun Jeong; Taewon Park; Jong Hwan Ko; Jungrae Kim

doi:10.1109/LCA.2022.3177419

Multi-Prediction Compression: An Efficient and Scalable Memory Compression Framework for GP-GPU

Hoyong Jin
, Donghun Jeong
, Taewon Park
, Jong Hwan Ko
, Jungrae Kim

Department of Electronic and Electrical Engineering

Sungkyunkwan University

Research output: Contribution to journal › Article › peer-review

5 Scopus citations

Abstract

Data-intensive applications and throughput-oriented processors demand more memory bandwidth. Memory compression can provide more data beyond physical limits, yet new data types and smaller block sizes are challenging. This paper presents a novel and lightweight memory compression framework, Multi-Prediction Compression (MPC), to increase the effective memory bandwidth. Based on multiple prediction models and data-driven algorithm tuning, MPC can provide 31.7% better compression than state-of-the-art (SOTA) algorithms for 32B blocks. Moreover, MPC is hardware-friendly and scalable to support a growing number of data patterns.

Original language	English
Pages (from-to)	37-40
Number of pages	4
Journal	IEEE Computer Architecture Letters
Volume	21
Issue number	2
DOIs	https://doi.org/10.1109/LCA.2022.3177419
State	Published - 2022

Keywords

data compaction and compression
Graphics processors
memory hierarchy

Access to Document

10.1109/LCA.2022.3177419

Cite this

@article{57de8293321a4d2b910212f3612f2c36,

title = "Multi-Prediction Compression: An Efficient and Scalable Memory Compression Framework for GP-GPU",

abstract = "Data-intensive applications and throughput-oriented processors demand more memory bandwidth. Memory compression can provide more data beyond physical limits, yet new data types and smaller block sizes are challenging. This paper presents a novel and lightweight memory compression framework, Multi-Prediction Compression (MPC), to increase the effective memory bandwidth. Based on multiple prediction models and data-driven algorithm tuning, MPC can provide 31.7\% better compression than state-of-the-art (SOTA) algorithms for 32B blocks. Moreover, MPC is hardware-friendly and scalable to support a growing number of data patterns.",

keywords = "data compaction and compression, Graphics processors, memory hierarchy",

author = "Hoyong Jin and Donghun Jeong and Taewon Park and Ko, \{Jong Hwan\} and Jungrae Kim",

note = "Publisher Copyright: {\textcopyright} 2002-2011 IEEE.",

year = "2022",

doi = "10.1109/LCA.2022.3177419",

language = "English",

volume = "21",

pages = "37--40",

journal = "IEEE Computer Architecture Letters",

issn = "1556-6056",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

number = "2",

}

TY - JOUR

T1 - Multi-Prediction Compression

T2 - An Efficient and Scalable Memory Compression Framework for GP-GPU

AU - Jin, Hoyong

AU - Jeong, Donghun

AU - Park, Taewon

AU - Ko, Jong Hwan

AU - Kim, Jungrae

PY - 2022

Y1 - 2022

N2 - Data-intensive applications and throughput-oriented processors demand more memory bandwidth. Memory compression can provide more data beyond physical limits, yet new data types and smaller block sizes are challenging. This paper presents a novel and lightweight memory compression framework, Multi-Prediction Compression (MPC), to increase the effective memory bandwidth. Based on multiple prediction models and data-driven algorithm tuning, MPC can provide 31.7% better compression than state-of-the-art (SOTA) algorithms for 32B blocks. Moreover, MPC is hardware-friendly and scalable to support a growing number of data patterns.

AB - Data-intensive applications and throughput-oriented processors demand more memory bandwidth. Memory compression can provide more data beyond physical limits, yet new data types and smaller block sizes are challenging. This paper presents a novel and lightweight memory compression framework, Multi-Prediction Compression (MPC), to increase the effective memory bandwidth. Based on multiple prediction models and data-driven algorithm tuning, MPC can provide 31.7% better compression than state-of-the-art (SOTA) algorithms for 32B blocks. Moreover, MPC is hardware-friendly and scalable to support a growing number of data patterns.

KW - data compaction and compression

KW - Graphics processors

KW - memory hierarchy

UR - https://www.scopus.com/pages/publications/85130779306

U2 - 10.1109/LCA.2022.3177419

DO - 10.1109/LCA.2022.3177419

M3 - Article

AN - SCOPUS:85130779306

SN - 1556-6056

VL - 21

SP - 37

EP - 40

JO - IEEE Computer Architecture Letters

JF - IEEE Computer Architecture Letters

IS - 2

ER -

Multi-Prediction Compression: An Efficient and Scalable Memory Compression Framework for GP-GPU

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this