Skip to main navigation Skip to search Skip to main content

A Computational Model for Tensor Core Units

  • University of Padua
  • Free University of Bozen-Bolzano

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

13 Scopus citations

Abstract

To respond to the need for efficient training and inference of deep neural networks, a plethora of domain-specific architectures have been introduced, such as Google Tensor Processing Units and NVIDIA Tensor Cores. A common feature of these architectures is the design for efficiently computing a dense matrix product of a given small size. In order to broaden the class of algorithms that exploit these systems, we propose a computational model, named the TCU model, that captures the ability to natively multiply small matrices. We then use the TCU model for designing fast algorithms for several problems, including dense and sparse matrix multiplication and the Discrete Fourier Transform. We finally highlight a relation between the TCU model and the external memory model.

Original languageEnglish
Title of host publicationSPAA 2020 - Proceedings of the 32nd ACM Symposium on Parallelism in Algorithms and Architectures
PublisherAssociation for Computing Machinery
Pages519-521
Number of pages3
ISBN (Electronic)9781450369350
DOIs
StatePublished - Jul 6 2020
Event32nd ACM Symposium on Parallelism in Algorithms and Architectures, SPAA 2020 - Virtual, Online, United States
Duration: Jul 15 2020Jul 17 2020

Publication series

NameAnnual ACM Symposium on Parallelism in Algorithms and Architectures

Conference

Conference32nd ACM Symposium on Parallelism in Algorithms and Architectures, SPAA 2020
Country/TerritoryUnited States
CityVirtual, Online
Period07/15/2007/17/20

Keywords

  • computational model
  • efficient algorithms
  • graph problems
  • hardware accelerators
  • linear algebra
  • tensor core

Fingerprint

Dive into the research topics of 'A Computational Model for Tensor Core Units'. Together they form a unique fingerprint.

Cite this