NCCL: NVIDIA Collective Communications Library

Fields

id

sources/src5-nccl-github

URL

https://github.com/NVIDIA/nccl

Publisher

NVIDIA

Kind

repo

Published

2016-01-01

Accessed

2026-10-04

notes

NCCL provides multi-GPU and multi-node collective communication primitives. Key for tensor parallel and pipeline parallel inference. Supports all-rereduce, all-gather, reduce-scatter, broadcast. Used by vLLM, TensorRT-LLM, SGLang for distributed inference.

Notes

NCCL provides multi-GPU and multi-node collective communication primitives. Key for tensor parallel and pipeline parallel inference. Supports all-rereduce, all-gather, reduce-scatter, broadcast. Used by vLLM, TensorRT-LLM, SGLang for distributed inference.

Cross-references

This record references no other records.