國立虎尾科技大學 |

Practical Tools for Reliable and Reproducible Graphics Processing Unit (GPU) Computations /

紀錄類型:	書目-語言資料,印刷品 : Monograph/item
正題名/作者:	Practical Tools for Reliable and Reproducible Graphics Processing Unit (GPU) Computations // Xinyi Li.
作者:	Li, Xinyi,
面頁冊數:	1 electronic resource (92 pages)
附註:	Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
Contained By:	Dissertations Abstracts International86-05B.
標題:	Computer science. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31327669
ISBN:	9798342716659

Practical Tools for Reliable and Reproducible Graphics Processing Unit (GPU) Computations /
Li, Xinyi,

Practical Tools for Reliable and Reproducible Graphics Processing Unit (GPU) Computations /Xinyi Li. - 1 electronic resource (92 pages)

Source: Dissertations Abstracts International, Volume: 86-05, Section: B.

This dissertation aims to address the correctness and reproducibility challenges in Graphical Process Unit (GPU)-accelerated applications, which are becoming increasingly prevalent in High-Performance Computing (HPC) and Artificial Intelligence (AI) domains. The growing use of GPUs as accelerators has introduced new challenges in ensuring the accuracy and consistency of numerical computations. One major source of these challenges is the inherent nature of floating-point arithmetic, which is heavily used in GPU-accelerated applications. Floating-point arithmetic is prone to numerical errors that can accumulate and lead to incorrect or inconsistent results. The massive parallelism of GPUs and the use of reduced precision formats exacerbate these challenges, making it difficult to detect and correct floating-point errors. Moreover, the lack of standardization in reduced precision formats across different GPU vendors poses significant challenges in maintaining the reliability and portability of GPU programs.This dissertation focuses on developing techniques and tools for floating-point error detection and tuning in GPU programs. The proposed work includes a tool called GPU-Floating Point Exception (GPU-FPX) that detects floating-point exceptions on Nvidia GPUs, and a test suite called Feature Targeted Tests for Numerical-libraries (FTTN) that identifies the numerical behaviors of Advanced Micro Devices (AMD) and Nvidia GPUs. These methodologies and tools aim to provide developers with practical solutions for understanding and addressing the challenges of numerical accuracy in GPU programming, ultimately enabling the creation of more robust and reliable GPU-accelerated applications across a wide range of domains, from scientific simulations to deep learning.

English

ISBN: 9798342716659Subjects--Topical Terms:

573171
Computer science.
Subjects--Index Terms:

Correctness

Practical Tools for Reliable and Reproducible Graphics Processing Unit (GPU) Computations /
LDR:03250nam a22004093i 4500 001 1157812
005 20250603111419.5
006 m o d
007 cr|nu||||||||
008 250804s2024 miu||||||m |||||||eng d
020 $a 9798342716659
035 $a (MiAaPQD)AAI31327669
035 $a AAI31327669
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Li, Xinyi, $e author. $3 1484087
245 1 0 $a Practical Tools for Reliable and Reproducible Graphics Processing Unit (GPU) Computations / $c Xinyi Li.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (92 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
500 $a Advisors: Gopalakrishnan, Ganesh Committee members: Hall, Mary W.; Sadayappan, Ponnuswamy; Li, Ang; Laguna, Ignacio.
502 $b Ph.D. $c The University of Utah $d 2024.
520 $a This dissertation aims to address the correctness and reproducibility challenges in Graphical Process Unit (GPU)-accelerated applications, which are becoming increasingly prevalent in High-Performance Computing (HPC) and Artificial Intelligence (AI) domains. The growing use of GPUs as accelerators has introduced new challenges in ensuring the accuracy and consistency of numerical computations. One major source of these challenges is the inherent nature of floating-point arithmetic, which is heavily used in GPU-accelerated applications. Floating-point arithmetic is prone to numerical errors that can accumulate and lead to incorrect or inconsistent results. The massive parallelism of GPUs and the use of reduced precision formats exacerbate these challenges, making it difficult to detect and correct floating-point errors. Moreover, the lack of standardization in reduced precision formats across different GPU vendors poses significant challenges in maintaining the reliability and portability of GPU programs.This dissertation focuses on developing techniques and tools for floating-point error detection and tuning in GPU programs. The proposed work includes a tool called GPU-Floating Point Exception (GPU-FPX) that detects floating-point exceptions on Nvidia GPUs, and a test suite called Feature Targeted Tests for Numerical-libraries (FTTN) that identifies the numerical behaviors of Advanced Micro Devices (AMD) and Nvidia GPUs. These methodologies and tools aim to provide developers with practical solutions for understanding and addressing the challenges of numerical accuracy in GPU programming, ultimately enabling the creation of more robust and reliable GPU-accelerated applications across a wide range of domains, from scientific simulations to deep learning.
546 $a English
590 $a School code: 0240
650 4 $a Computer science. $3 573171
653 $a Correctness
653 $a Floating-point
653 $a Advanced Micro Devices
653 $a Graphics Processing Unit
690 $a 0984
690 $a 0800
710 2 $a The University of Utah. $b School of Computing. $3 1188323
720 1 $a Gopalakrishnan, Ganesh $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-05B.
790 $a 0240
791 $a Ph.D.
792 $a 2024
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31327669