Solving sparse linear systems with approximate inverse preconditioners on analog devices

Vasileios Kalantzis; Anshul Gupta; Lior Horesh; Tomasz Nowicki; Mark S. Squillante; Chai Wah Wu; Tayfun Gokmen; Haim Avron

doi:10.1109/HPEC49654.2021.9622816

Solving sparse linear systems with approximate inverse preconditioners on analog devices

Vasileios Kalantzis, Anshul Gupta, Lior Horesh, Tomasz Nowicki, Mark S. Squillante, Chai Wah Wu, Tayfun Gokmen, Haim Avron

School of Mathematical Sciences

IBM

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

2 Scopus citations

Abstract

Sparse linear system solvers are computationally expensive kernels that lie at the heart of numerous applications. This paper proposes a preconditioning framework that combines approximate inverses with stationary iterations to substantially reduce the time and energy requirements of this task by utilizing a hybrid architecture that combines conventional digital microprocessors with analog crossbar array accelerators. Our analysis and experiments with a simulator for analog hardware show that an order of magnitude speedup is readily attainable despite the noise in analog computations.

Original language	English
Title of host publication	2021 IEEE High Performance Extreme Computing Conference, HPEC 2021
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781665423694
DOIs	https://doi.org/10.1109/HPEC49654.2021.9622816
State	Published - 2021
Event	2021 IEEE High Performance Extreme Computing Conference, HPEC 2021 - Virtual, Online, United States Duration: 20 Sep 2021 → 24 Sep 2021

Publication series

Name	2021 IEEE High Performance Extreme Computing Conference, HPEC 2021

Conference

Conference	2021 IEEE High Performance Extreme Computing Conference, HPEC 2021
Country/Territory	United States
City	Virtual, Online
Period	20/09/21 → 24/09/21

Keywords

Richardson iteration
analog crossbar arrays
approximate inverse pre-conditioners

Access to Document

10.1109/HPEC49654.2021.9622816

Cite this

Kalantzis, V., Gupta, A., Horesh, L., Nowicki, T., Squillante, M. S., Wah Wu, C., Gokmen, T., & Avron, H. (2021). Solving sparse linear systems with approximate inverse preconditioners on analog devices. In 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021 (2021 IEEE High Performance Extreme Computing Conference, HPEC 2021). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/HPEC49654.2021.9622816

@inproceedings{4d406cbe558843208dcb94823d9cc632,

title = "Solving sparse linear systems with approximate inverse preconditioners on analog devices",

abstract = "Sparse linear system solvers are computationally expensive kernels that lie at the heart of numerous applications. This paper proposes a preconditioning framework that combines approximate inverses with stationary iterations to substantially reduce the time and energy requirements of this task by utilizing a hybrid architecture that combines conventional digital microprocessors with analog crossbar array accelerators. Our analysis and experiments with a simulator for analog hardware show that an order of magnitude speedup is readily attainable despite the noise in analog computations.",

keywords = "Richardson iteration, analog crossbar arrays, approximate inverse pre-conditioners",

author = "Vasileios Kalantzis and Anshul Gupta and Lior Horesh and Tomasz Nowicki and Squillante, {Mark S.} and {Wah Wu}, Chai and Tayfun Gokmen and Haim Avron",

note = "Publisher Copyright: {\textcopyright} 2021 IEEE.; 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021 ; Conference date: 20-09-2021 Through 24-09-2021",

year = "2021",

doi = "10.1109/HPEC49654.2021.9622816",

language = "אנגלית",

series = "2021 IEEE High Performance Extreme Computing Conference, HPEC 2021",

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

booktitle = "2021 IEEE High Performance Extreme Computing Conference, HPEC 2021",

address = "ארצות הברית",

}

Kalantzis, V, Gupta, A, Horesh, L, Nowicki, T, Squillante, MS, Wah Wu, C, Gokmen, T & Avron, H 2021, Solving sparse linear systems with approximate inverse preconditioners on analog devices. in 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021. 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021, Institute of Electrical and Electronics Engineers Inc., 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021, Virtual, Online, United States, 20/09/21. https://doi.org/10.1109/HPEC49654.2021.9622816

Solving sparse linear systems with approximate inverse preconditioners on analog devices. / Kalantzis, Vasileios; Gupta, Anshul; Horesh, Lior et al.
2021 IEEE High Performance Extreme Computing Conference, HPEC 2021. Institute of Electrical and Electronics Engineers Inc., 2021. (2021 IEEE High Performance Extreme Computing Conference, HPEC 2021).

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

TY - GEN

T1 - Solving sparse linear systems with approximate inverse preconditioners on analog devices

AU - Kalantzis, Vasileios

AU - Gupta, Anshul

AU - Horesh, Lior

AU - Nowicki, Tomasz

AU - Squillante, Mark S.

AU - Wah Wu, Chai

AU - Gokmen, Tayfun

AU - Avron, Haim

PY - 2021

Y1 - 2021

N2 - Sparse linear system solvers are computationally expensive kernels that lie at the heart of numerous applications. This paper proposes a preconditioning framework that combines approximate inverses with stationary iterations to substantially reduce the time and energy requirements of this task by utilizing a hybrid architecture that combines conventional digital microprocessors with analog crossbar array accelerators. Our analysis and experiments with a simulator for analog hardware show that an order of magnitude speedup is readily attainable despite the noise in analog computations.

AB - Sparse linear system solvers are computationally expensive kernels that lie at the heart of numerous applications. This paper proposes a preconditioning framework that combines approximate inverses with stationary iterations to substantially reduce the time and energy requirements of this task by utilizing a hybrid architecture that combines conventional digital microprocessors with analog crossbar array accelerators. Our analysis and experiments with a simulator for analog hardware show that an order of magnitude speedup is readily attainable despite the noise in analog computations.

KW - Richardson iteration

KW - analog crossbar arrays

KW - approximate inverse pre-conditioners

UR - http://www.scopus.com/inward/record.url?scp=85123471491&partnerID=8YFLogxK

U2 - 10.1109/HPEC49654.2021.9622816

DO - 10.1109/HPEC49654.2021.9622816

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:85123471491

T3 - 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021

BT - 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021

Y2 - 20 September 2021 through 24 September 2021

ER -

Kalantzis V, Gupta A, Horesh L, Nowicki T, Squillante MS, Wah Wu C et al. Solving sparse linear systems with approximate inverse preconditioners on analog devices. In 2021 IEEE High Performance Extreme Computing Conference, HPEC 2021. Institute of Electrical and Electronics Engineers Inc. 2021. (2021 IEEE High Performance Extreme Computing Conference, HPEC 2021). doi: 10.1109/HPEC49654.2021.9622816

Solving sparse linear systems with approximate inverse preconditioners on analog devices

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this