Exascale Deep Learning for Climate Analytics

We extract pixel-level masks of extreme weather patterns using variants of Tiramisu and DeepLabv3+ neural networks. We describe improvements to the software frameworks, input pipeline, and the network training algorithms necessary to efficiently scale deep learning on the Piz Daint and Summit system...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:SC18: International Conference for High Performance Computing, Networking, Storage and Analysis S. 649 - 660
Hauptverfasser: Kurth, Thorsten, Treichler, Sean, Romero, Joshua, Mudigonda, Mayur, Luehr, Nathan, Phillips, Everett, Mahesh, Ankur, Matheson, Michael, Deslippe, Jack, Fatica, Massimiliano, Prabhat, Prabhat, Houston, Michael
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 01.11.2018
Schlagworte:
Computational modeling
Computer architecture
Convolutional codes
Deep learning
Meteorology
Technological innovation
Training
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Abstract We extract pixel-level masks of extreme weather patterns using variants of Tiramisu and DeepLabv3+ neural networks. We describe improvements to the software frameworks, input pipeline, and the network training algorithms necessary to efficiently scale deep learning on the Piz Daint and Summit systems. The Tiramisu network scales to 5300 P100 GPUs with a sustained throughput of 21.0 PF/s and parallel efficiency of 79.0%. DeepLabv3+ scales up to 27360 V100 GPUs with a sustained throughput of 325.8 PF/s and a parallel efficiency of 90.7% in single precision. By taking advantage of the FP16 Tensor Cores, a half-precision version of the DeepLabv3+ network achieves a peak and sustained throughput of 1.13 EF/s and 999.0 PF/s respectively.
AbstractList We extract pixel-level masks of extreme weather patterns using variants of Tiramisu and DeepLabv3+ neural networks. We describe improvements to the software frameworks, input pipeline, and the network training algorithms necessary to efficiently scale deep learning on the Piz Daint and Summit systems. The Tiramisu network scales to 5300 P100 GPUs with a sustained throughput of 21.0 PF/s and parallel efficiency of 79.0%. DeepLabv3+ scales up to 27360 V100 GPUs with a sustained throughput of 325.8 PF/s and a parallel efficiency of 90.7% in single precision. By taking advantage of the FP16 Tensor Cores, a half-precision version of the DeepLabv3+ network achieves a peak and sustained throughput of 1.13 EF/s and 999.0 PF/s respectively.
Author Romero, Joshua
Fatica, Massimiliano
Deslippe, Jack
Prabhat, Prabhat
Kurth, Thorsten
Phillips, Everett
Treichler, Sean
Mudigonda, Mayur
Mahesh, Ankur
Houston, Michael
Luehr, Nathan
Matheson, Michael
Author_xml – sequence: 1
  givenname: Thorsten
  surname: Kurth
  fullname: Kurth, Thorsten
– sequence: 2
  givenname: Sean
  surname: Treichler
  fullname: Treichler, Sean
– sequence: 3
  givenname: Joshua
  surname: Romero
  fullname: Romero, Joshua
– sequence: 4
  givenname: Mayur
  surname: Mudigonda
  fullname: Mudigonda, Mayur
– sequence: 5
  givenname: Nathan
  surname: Luehr
  fullname: Luehr, Nathan
– sequence: 6
  givenname: Everett
  surname: Phillips
  fullname: Phillips, Everett
– sequence: 7
  givenname: Ankur
  surname: Mahesh
  fullname: Mahesh, Ankur
– sequence: 8
  givenname: Michael
  surname: Matheson
  fullname: Matheson, Michael
– sequence: 9
  givenname: Jack
  surname: Deslippe
  fullname: Deslippe, Jack
– sequence: 10
  givenname: Massimiliano
  surname: Fatica
  fullname: Fatica, Massimiliano
– sequence: 11
  givenname: Prabhat
  surname: Prabhat
  fullname: Prabhat, Prabhat
– sequence: 12
  givenname: Michael
  surname: Houston
  fullname: Houston, Michael
BookMark eNotzrFOwzAQgGEjwdAWxk5d_AAk-HyJ4xurUApSJAZgrs7JuYoU3CrJQN-eSjD926d_qW7TKYlSazA5gKGnjzq3BnxujCmLG7WEEr3z6AtaqMfdD08tD6KfRc66ER5Tn446nkZdD_03z6K3iYfL3LfTvbqLPEzy8N-V-nrZfdavWfO-f6u3TcZQ-jmrPHYxdOzAMkIgInQRQwc2Bg8WMVBhWrKBDAQgJ4HJsg9QmcoV4nClNn9uLyKH83jdGC8H71xZXa1fyhM8og
CODEN IEEPAD
ContentType Conference Proceeding
DBID 6IE
6IL
CBEJK
RIE
RIL
DOI 10.1109/SC.2018.00054
DatabaseName IEEE Electronic Library (IEL) Conference Proceedings
IEEE Proceedings Order Plan All Online (POP All Online) 1998-present by volume
IEEE Xplore All Conference Proceedings
IEEE Electronic Library (IEL)
IEEE Proceedings Order Plans (POP All) 1998-Present
DatabaseTitleList
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Electronic Library (IEL)
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
EISBN 1538683849
9781538683842
EndPage 660
ExternalDocumentID 8665799
Genre orig-research
GroupedDBID 6IE
6IL
CBEJK
RIE
RIL
ID FETCH-LOGICAL-a158t-783dfbda612a31b99936f3bd12fb81233b940c92b901b196eba92a8b170764e63
IEDL.DBID RIE
IngestDate Thu Jun 29 18:39:01 EDT 2023
IsPeerReviewed false
IsScholarly false
Language English
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-a158t-783dfbda612a31b99936f3bd12fb81233b940c92b901b196eba92a8b170764e63
PageCount 12
ParticipantIDs ieee_primary_8665799
PublicationCentury 2000
PublicationDate 2018-Nov
PublicationDateYYYYMMDD 2018-11-01
PublicationDate_xml – month: 11
  year: 2018
  text: 2018-Nov
PublicationDecade 2010
PublicationTitle SC18: International Conference for High Performance Computing, Networking, Storage and Analysis
PublicationTitleAbbrev SC
PublicationYear 2018
Publisher IEEE
Publisher_xml – name: IEEE
Score 2.239504
Snippet We extract pixel-level masks of extreme weather patterns using variants of Tiramisu and DeepLabv3+ neural networks. We describe improvements to the software...
SourceID ieee
SourceType Publisher
StartPage 649
SubjectTerms Computational modeling
Computer architecture
Convolutional codes
Deep learning
Meteorology
Technological innovation
Training
Title Exascale Deep Learning for Climate Analytics
URI https://ieeexplore.ieee.org/document/8665799
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3LagIxFL2odNFVW7T0TRZdmjqTzOSxtkpXIrQFd5LHnSIUFV_085tkBt10011IFiEHwjkk95wL8MyY504YR5nXihY6M9SUvqClcF7roBDCYmo2IScTNZvpaQv6Ry8MIqbiM3yJw_SX71duH5_KBjGbTWrdhraUovZqnWIzB-_DWKkVSyOzFO5_apaSuGJ88b9dLqF3Mt2R6ZFOrqCFyy70Rz9mG2BE8oq4Jk0a6hcJUpMMvxdBbiJJuSIxbbkHn-PRx_CNNg0OqMlLtaNScV9Zb4LKMDy3QatxUXHrc1bZQLyc24CW08wG0rbhqqA1mhllc5lJUaDg19BZrpZ4A6QwRjLkzjOBhXDMonaZ8lhl3jIp5C1040nn6zrDYt4c8u7v6Xs4j1DWnrsH6Ow2e3yEM3fYLbabpwT8L9W7hgA
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV1NSwMxEB1qFfSk0orf5uCxsbvJbrI515aKtRSs0FvJx6wUpC39EH--SVraixdvITmEmRDeI5n3BuCRMcet0JYypwqaqURTnbuM5sI6pTxD8Iux2YTs94vRSA0q0NhpYRAxFp_hUxjGv3w3s-vwVNYM3mxSqQM4zD2OJhu11t44s_neCrVaoTgyifb--3YpES06p__b5wzqe9kdGewA5RwqOK1Bo_2jlz6RSJ4R52Trh_pJPNkkra-JJ5xIorNI8Fuuw0enPWx16bbFAdVpXqyoLLgrjdOeZ2ieGs_WuCi5cSkrjYdezo3Pl1XMeNg2_rKg0YrpwqQykSJDwS-gOp1N8RJIprVkyK1jAjNhmUFlk8JhmTjDpJBXUAuRjucbF4vxNsjrv6cf4Lg7fOuNey_91xs4CWndKPBuobparPEOjuz3arJc3MdD-AUKz4lN
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=proceeding&rft.title=SC18%3A+International+Conference+for+High+Performance+Computing%2C+Networking%2C+Storage+and+Analysis&rft.atitle=Exascale+Deep+Learning+for+Climate+Analytics&rft.au=Kurth%2C+Thorsten&rft.au=Treichler%2C+Sean&rft.au=Romero%2C+Joshua&rft.au=Mudigonda%2C+Mayur&rft.date=2018-11-01&rft.pub=IEEE&rft.spage=649&rft.epage=660&rft_id=info:doi/10.1109%2FSC.2018.00054&rft.externalDocID=8665799