Harnessing GPU Tensor Cores for Fast FP16 Arithmetic to Speed up Mixed-Precision Iterative Refinement Solvers

Azzam  Haidar; Stanimire Tomov; Jack Dongarra; Nicholas Higham

doi:10.1109/SC.2018.00050

Harnessing GPU Tensor Cores for Fast FP16 Arithmetic to Speed up Mixed-Precision Iterative Refinement Solvers

Azzam Haidar, Stanimire Tomov, Jack Dongarra, Nicholas Higham

Software Systems

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

488 Downloads (Pure)

Abstract

Low-precision floating-point arithmetic is a powerful tool for accelerating scientific computing applications, especially those in artificial intelligence. Here, we present an investigation showing that other high-performance computing
(HPC) applications can also harness this power. Specifically, we use the general HPC problem, Ax = b, where A is a large dense matrix, and a double precision (FP64) solution is needed for accuracy. Our approach is based on mixed-precision (FP16!FP64) iterative refinement, and we generalize and extend prior advances into a framework, for which we develop architecture-specific algorithms and highly tuned implementations. These new methods show how using half-precision Tensor Cores (FP16-TC) for the arithmetic can provide up to 4 speedup. This is due to the performance boost that the FP16-TC provide as well as to the improved accuracy over the classical FP16 arithmetic that is obtained

Original language	English
Title of host publication	SC '18: Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis
Publisher	IEEE
Pages	1-11
Volume	0
DOIs	https://doi.org/10.1109/SC.2018.00050
Publication status	Published - 11 Nov 2018

Keywords

FP16 Arithmetic
Half Precision
Mixed Precision Solvers
Iterative Refinement Computation
GPU Computing
Linear algebra

Access to Document

10.1109/SC.2018.00050

mainAccepted author manuscript, 1.04 MB

https://dl.acm.org/citation.cfm?id=3291719

Cite this

@inproceedings{accab98e4e5a47419aaaea1701674899,

title = "Harnessing GPU Tensor Cores for Fast FP16 Arithmetic to Speed up Mixed-Precision Iterative Refinement Solvers",

abstract = "Low-precision floating-point arithmetic is a powerful tool for accelerating scientific computing applications, especially those in artificial intelligence. Here, we present an investigation showing that other high-performance computing(HPC) applications can also harness this power. Specifically, we use the general HPC problem, Ax = b, where A is a large dense matrix, and a double precision (FP64) solution is needed for accuracy. Our approach is based on mixed-precision (FP16!FP64) iterative refinement, and we generalize and extend prior advances into a framework, for which we develop architecture-specific algorithms and highly tuned implementations. These new methods show how using half-precision Tensor Cores (FP16-TC) for the arithmetic can provide up to 4 speedup. This is due to the performance boost that the FP16-TC provide as well as to the improved accuracy over the classical FP16 arithmetic that is obtained",

keywords = "FP16 Arithmetic, Half Precision, Mixed Precision Solvers, Iterative Refinement Computation, GPU Computing, Linear algebra",

author = "Azzam Haidar and Stanimire Tomov and Jack Dongarra and Nicholas Higham",

year = "2018",

month = nov,

day = "11",

doi = "10.1109/SC.2018.00050",

language = "English",

volume = "0",

pages = "1--11",

booktitle = "SC '18: Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis",

publisher = "IEEE",

address = "United States",

}

Harnessing GPU Tensor Cores for Fast FP16 Arithmetic to Speed up Mixed-Precision Iterative Refinement Solvers. / Haidar, Azzam ; Tomov, Stanimire; Dongarra, Jack et al.

SC '18: Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis. Vol. 0 IEEE, 2018. p. 1-11 47.

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

TY - GEN

T1 - Harnessing GPU Tensor Cores for Fast FP16 Arithmetic to Speed up Mixed-Precision Iterative Refinement Solvers

AU - Haidar, Azzam

AU - Tomov, Stanimire

AU - Dongarra, Jack

AU - Higham, Nicholas

PY - 2018/11/11

Y1 - 2018/11/11

N2 - Low-precision floating-point arithmetic is a powerful tool for accelerating scientific computing applications, especially those in artificial intelligence. Here, we present an investigation showing that other high-performance computing(HPC) applications can also harness this power. Specifically, we use the general HPC problem, Ax = b, where A is a large dense matrix, and a double precision (FP64) solution is needed for accuracy. Our approach is based on mixed-precision (FP16!FP64) iterative refinement, and we generalize and extend prior advances into a framework, for which we develop architecture-specific algorithms and highly tuned implementations. These new methods show how using half-precision Tensor Cores (FP16-TC) for the arithmetic can provide up to 4 speedup. This is due to the performance boost that the FP16-TC provide as well as to the improved accuracy over the classical FP16 arithmetic that is obtained

AB - Low-precision floating-point arithmetic is a powerful tool for accelerating scientific computing applications, especially those in artificial intelligence. Here, we present an investigation showing that other high-performance computing(HPC) applications can also harness this power. Specifically, we use the general HPC problem, Ax = b, where A is a large dense matrix, and a double precision (FP64) solution is needed for accuracy. Our approach is based on mixed-precision (FP16!FP64) iterative refinement, and we generalize and extend prior advances into a framework, for which we develop architecture-specific algorithms and highly tuned implementations. These new methods show how using half-precision Tensor Cores (FP16-TC) for the arithmetic can provide up to 4 speedup. This is due to the performance boost that the FP16-TC provide as well as to the improved accuracy over the classical FP16 arithmetic that is obtained

KW - FP16 Arithmetic

KW - Half Precision

KW - Mixed Precision Solvers

KW - Iterative Refinement Computation

KW - GPU Computing

KW - Linear algebra

U2 - 10.1109/SC.2018.00050

DO - 10.1109/SC.2018.00050

M3 - Conference contribution

VL - 0

SP - 1

EP - 11

BT - SC '18: Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis

PB - IEEE

ER -

Harnessing GPU Tensor Cores for Fast FP16 Arithmetic to Speed up Mixed-Precision Iterative Refinement Solvers

Abstract

Keywords

Access to Document

Fingerprint

Cite this