NVIDIA Collective Communications Library

URI: https://gptkb.org/entity/NVIDIA_Collective_Communications_Library
GPTKB entity
Statements (38)
Predicate Object
gptkbp:instanceOf gptkb:software
gptkbp:abbreviation gptkb:NCCL
gptkbp:category deep learning
distributed computing
parallel computing
gptkbp:developer gptkb:NVIDIA
gptkbp:feature low latency
scalability
high bandwidth
peer-to-peer communication
multi-process support
gptkbp:firstReleased 2016
https://www.w3.org/2000/01/rdf-schema#label NVIDIA Collective Communications Library
gptkbp:integratesWith gptkb:TensorFlow
gptkb:MXNet
gptkb:PyTorch
gptkb:Horovod
gptkbp:language gptkb:C++
C
gptkbp:latestReleaseVersion NCCL 2.x
gptkbp:license gptkb:BSD-3-Clause
gptkbp:operatingSystem gptkb:Linux
gptkbp:optimizedFor gptkb:NVIDIA_GPUs
gptkbp:provides gptkb:broadcaster
reduce
all-gather
all-reduce
reduce-scatter
collective communication primitives
gptkbp:purpose multi-GPU communication
multi-node communication
gptkbp:supports gptkb:CUDA
GPUs
deep learning frameworks
gptkbp:usedIn distributed deep learning
gptkbp:website https://developer.nvidia.com/nccl
gptkbp:bfsParent gptkb:NCCL
gptkbp:bfsLayer 7