Processing

Please wait...

PATENTSCOPE will be unavailable a few hours for maintenance reason on Tuesday 25.01.2022 at 9:00 AM CET
Settings

Settings

Goto Application

1. WO2021197657 - CONTROL OF DATA TRANSFER BETWEEN PROCESSING NODES

Publication Number WO/2021/197657
Publication Date 07.10.2021
International Application No. PCT/EP2020/087568
International Filing Date 22.12.2020
IPC
G06F 15/173 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
FELECTRIC DIGITAL DATA PROCESSING
15Digital computers in general; Data processing equipment in general
16Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
163Interprocessor communication
173using an interconnection network, e.g. matrix, shuffle, pyramid, star or snowflake
G06N 3/08 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
G06F 9/54 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
FELECTRIC DIGITAL DATA PROCESSING
9Arrangements for program control, e.g. control units
06using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
46Multiprogramming arrangements
54Interprogram communication
CPC
G06F 15/17318
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
FELECTRIC DIGITAL DATA PROCESSING
15Digital computers in general
16Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
163Interprocessor communication
173using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
17306Intercommunication techniques
17318Parallel communications techniques, e.g. gather, scatter, reduce, roadcast, multicast, all to all
G06F 15/17325
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
FELECTRIC DIGITAL DATA PROCESSING
15Digital computers in general
16Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
163Interprocessor communication
173using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
17306Intercommunication techniques
17325Synchronisation; Hardware support therefor
G06F 9/52
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
FELECTRIC DIGITAL DATA PROCESSING
9Arrangements for program control, e.g. control units
06using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
46Multiprogramming arrangements
52Program synchronisation; Mutual exclusion, e.g. by means of semaphores
G06F 9/522
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
FELECTRIC DIGITAL DATA PROCESSING
9Arrangements for program control, e.g. control units
06using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
46Multiprogramming arrangements
52Program synchronisation; Mutual exclusion, e.g. by means of semaphores
522Barrier synchronisation
G06F 9/542
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
FELECTRIC DIGITAL DATA PROCESSING
9Arrangements for program control, e.g. control units
06using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
46Multiprogramming arrangements
54Interprogram communication
542Event management; Broadcasting; Multicasting; Notifications
G06N 3/02
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
Applicants
  • GRAPHCORE LIMITED [GB]/[GB]
Inventors
  • CEVOLANI, Lorenzo
  • TSCHOPP, Fabian
  • TØRUDBAKKEN, Ola
Agents
  • KEELING, Thomas
Priority Data
16/928,78214.07.2020US
2004857.502.04.2020GB
2008390.304.06.2020GB
Publication Language English (en)
Filing Language English (EN)
Designated States
Title
(EN) CONTROL OF DATA TRANSFER BETWEEN PROCESSING NODES
(FR) COMMANDE DE TRANSFERT DE DONNÉES ENTRE NŒUDS DE TRAITEMENT
Abstract
(EN) A data processing system comprising a plurality of processing nodes, each comprising at least one memory configured to store an array of data items, wherein each of the plurality of processing nodes is configured to execute compute instructions during a compute phase and following a precompiled synchronisation barrier, enter at least one exchange phase. During the at least one exchange phase, a series of collective operations are carried out. Each processing node is configured to perform a reduce scatter collective in at least one first dimension. Using the results of the reduce scatter collective, each processing node performs an allreduce in a second dimension. The processing nodes then perform an all-gather collective in the at least one first dimension using the results of the allreduce.
(FR) Un système de traitement de données comprend une pluralité de nœuds de traitement, comprenant chacun au moins une mémoire configurée pour stocker un réseau d'éléments de données, chaque nœud de la pluralité de nœuds de traitement étant configuré pour exécuter des instructions de calcul pendant une phase de calcul et après une barrière de synchronisation précompilée, pour entrer au moins dans une phase d'échange. Pendant ladite phase d'échange, une série d'opérations collectives sont effectuées. Chaque nœud de traitement est configuré pour effectuer un collectif de diffusion de réduction dans au moins une première dimension. En utilisant les résultats du collectif de diffusion de réduction, chaque nœud de traitement effectue un algorithme allreduce dans une seconde dimension. Les nœuds de traitement effectuent ensuite un collectif all-gather dans ladite première dimension en utilisant les résultats de l'algorithme allreduce.
Latest bibliographic data on file with the International Bureau