kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
GPU Implementation of the Triple Tensor Decomposition
KTH, School of Electrical Engineering and Computer Science (EECS).
2024 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
GPU-implementering av Triple Tensor Decomposition (Swedish)
Abstract [en]

The triple decomposition of the velocity gradient tensor provides an analytical tool for fluid dynamics by which the flow can be separated into a sum of nonrotationally strained, shear, and rigidly rotating flows. Recently, the triple decomposition has again been shown to be efficiently computed via the Schur form of the correlation matrix.

In recent research, GPU-parallelized computing languages such as CUDA have also been applied to the study of computing large amounts of linear algebra data. By porting mathematical computational activities such as matrix operations from the CPU to the GPU, we reduce the time spent on the entire operation by massively parallelizing the computation.

In this article, we describe how this computational approach can be accelerated by means of a GPU-parallelized language, which can drastically improve computational efficiency when large amounts of data are available. And the results of the calculations are also visualized for easy comparison with the accuracy of existing CPU solutions.

The experimental process consists of three parts: first, the data is processed into Schur normalized matrices, then key information about the triple tensor decomposition is obtained by processing these matrices, and finally, the algorithm is tested for accuracy and precision by comparing it to currently available CPU solutions.

Abstract [sv]

Tredubbel nedbrytning av hastighetsgradienttensorn ger ett analytiskt verktyg för fluidmekanik genom vilket flödet kan separeras i en summa av icke-rotativt ansträngda, skjuvande och stelt roterande flöden. Nyligen har det visats att den tredubbla nedbrytningen effektivt kan beräknas genom Schurs form av korrelationsmatrisen.

I nyligen forskning har GPU-parallelliserade beräkningsspråk som CUDA också tillämpats på studier av beräkning av stora mängder linjär algebraisk data. Genom att flytta matematiska beräkningsaktiviteter såsom matrisoperationer från CPU till GPU minskar vi tiden som spenderas på hela operationen genom att massivt parallellisera beräkningen.

I den här artikeln beskriver vi hur detta beräkningsmetod kan accelereras med hjälp av ett GPU-parallelliserat språk, vilket kraftigt kan förbättra beräkningskapaciteten när stora mängder data är tillgängliga. Dessutom visualiseras resultaten av beräkningarna för enkel jämförelse med noggrannheten hos befintliga CPU-lösningar.

Den experimentella processen består av tre delar: först bearbetas data till Schur-normaliserade matriser, sedan erhålls nyckelinformation om den tredubbla tensornedbrytningen genom att bearbeta dessa matriser, och slutligen testas algoritmen för noggrannhet och precision genom att jämföra den med för närvarande tillgängliga CPU-lösningar.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology , 2024. , p. 19
Series
TRITA-EECS-EX ; 2024:424
Keywords [en]
GPU acceleration, High-performance computing, Triple tensor decomposition, Computational fluid dynamics
Keywords [sv]
GPU-acceleration, högpresterande databehandling, trippel tensor-dekomponering, beräkningsströmningsdynamik
National Category
Computer Sciences Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-351260OAI: oai:DiVA.org:kth-351260DiVA, id: diva2:1886756
Presentation
2024-06-11, via Zoom https://kth-se.zoom.us/j/5746215403, Stockholm, 13:00 (English)
Supervisors
Examiners
Available from: 2024-09-19 Created: 2024-08-04 Last updated: 2025-02-18Bibliographically approved

Open Access in DiVA

fulltext(1774 kB)105 downloads
File information
File name FULLTEXT01.pdfFile size 1774 kBChecksum SHA-512
5af14103d7580bf541e13a77d6b0c00f8aadb15d714464a619e7a5237555a90c5096bb689cea6080993a99553866ef6df98a452d45e87abd97073523e3950a18
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Computer SciencesComputer and Information Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 105 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 129 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf