Energy-Efficient Transprecision Techniques for Iterative Refinement

JunKyu Lee; Hans Vandierendonck; Dimitrios Nikolopoulos

Energy-Efficient Transprecision Techniques for Iterative Refinement

JunKyu Lee, Hans Vandierendonck, Dimitrios Nikolopoulos

School of Electronics, Electrical Engineering and Computer Science

Research output: Contribution to conference › Poster › peer-review

203 Downloads (Pure)

Abstract

This paper presents transprecision techniques for iterative refinement, which utilize various precision arithmetic dynamically according to numeric properties of the algorithm and computational latencies depending on precisions. The transprecision techniques were plugged into a mixed precision iterative refinement on an Intel Xeon E5-2650 2GHz core with MKL 2017 and XBLAS 1.0. The transprecision techniques brought further 2.0-3.4X speedups and 3.0-4.1X energy reductions to a mixed precision iterative refinement when double precision solution accuracy was required for forward error and a matrix size was ranged from 4K to 32K.

Original language	English
Publication status	Accepted - 01 Sept 2017
Event	Supercomputing'17 (SC17): International Conference on High Performance Computing, Networking, Storage and Analysis - Denver, United States Duration: 11 Nov 2017 → 17 Nov 2017

Conference

Conference	Supercomputing'17 (SC17): International Conference on High Performance Computing, Networking, Storage and Analysis
Abbreviated title	SC17
Country/Territory	United States
City	Denver
Period	11/11/2017 → 17/11/2017

Keywords

transprecision, mixed precision, iterative refinement

Access to Document

Energy-Efficient Transprecision Techniques for Iterative Refinement Poster
Copyright 2017 IEEE This work is made available online in accordance with the publisher’s policies. Please refer to any applicable terms of use of the publisher.
Licence: Unspecified
Energy-Efficient Transprecision Techniques for Iterative Refinement
Copyright 2017 IEEE This work is made available online in accordance with the publisher’s policies. Please refer to any applicable terms of use of the publisher.
Licence: Unspecified

Cite this

@conference{aa7c6130d12a403386daa9baa12c2406,

title = "Energy-Efficient Transprecision Techniques for Iterative Refinement",

abstract = "This paper presents transprecision techniques for iterative refinement, which utilize various precision arithmetic dynamically according to numeric properties of the algorithm and computational latencies depending on precisions. The transprecision techniques were plugged into a mixed precision iterative refinement on an Intel Xeon E5-2650 2GHz core with MKL 2017 and XBLAS 1.0. The transprecision techniques brought further 2.0-3.4X speedups and 3.0-4.1X energy reductions to a mixed precision iterative refinement when double precision solution accuracy was required for forward error and a matrix size was ranged from 4K to 32K.",

keywords = "transprecision, mixed precision, iterative refinement",

author = "JunKyu Lee and Hans Vandierendonck and Dimitrios Nikolopoulos",

year = "2017",

month = sep,

day = "1",

language = "English",

note = "Supercomputing'17 (SC17): International Conference on High Performance Computing, Networking, Storage and Analysis, SC17 ; Conference date: 11-11-2017 Through 17-11-2017",

}

Energy-Efficient Transprecision Techniques for Iterative Refinement. / Lee, JunKyu; Vandierendonck, Hans; Nikolopoulos, Dimitrios.
2017. Poster session presented at Supercomputing'17 (SC17): International Conference on High Performance Computing, Networking, Storage and Analysis, Denver, United States.

Research output: Contribution to conference › Poster › peer-review

TY - CONF

T1 - Energy-Efficient Transprecision Techniques for Iterative Refinement

AU - Lee, JunKyu

AU - Vandierendonck, Hans

AU - Nikolopoulos, Dimitrios

PY - 2017/9/1

Y1 - 2017/9/1

N2 - This paper presents transprecision techniques for iterative refinement, which utilize various precision arithmetic dynamically according to numeric properties of the algorithm and computational latencies depending on precisions. The transprecision techniques were plugged into a mixed precision iterative refinement on an Intel Xeon E5-2650 2GHz core with MKL 2017 and XBLAS 1.0. The transprecision techniques brought further 2.0-3.4X speedups and 3.0-4.1X energy reductions to a mixed precision iterative refinement when double precision solution accuracy was required for forward error and a matrix size was ranged from 4K to 32K.

AB - This paper presents transprecision techniques for iterative refinement, which utilize various precision arithmetic dynamically according to numeric properties of the algorithm and computational latencies depending on precisions. The transprecision techniques were plugged into a mixed precision iterative refinement on an Intel Xeon E5-2650 2GHz core with MKL 2017 and XBLAS 1.0. The transprecision techniques brought further 2.0-3.4X speedups and 3.0-4.1X energy reductions to a mixed precision iterative refinement when double precision solution accuracy was required for forward error and a matrix size was ranged from 4K to 32K.

KW - transprecision, mixed precision, iterative refinement

M3 - Poster

T2 - Supercomputing'17 (SC17): International Conference on High Performance Computing, Networking, Storage and Analysis

Y2 - 11 November 2017 through 17 November 2017

ER -

Energy-Efficient Transprecision Techniques for Iterative Refinement

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this