NVIDIA / nccl
git clone https://github.com/NVIDIA/nccl.git
v2.32.3-1 · 1mo ago
NVIDIA's C++/CUDA library of optimized collective communication primitives (all-reduce, all-gather, broadcast, etc.) for multi-GPU and multi-node systems.
Why you might care
Used as the communication backend by many deep learning frameworks for multi-GPU training. It targets high bandwidth over PCIe, NVLink, NVSwitch, InfiniBand, and TCP/IP sockets, and works in single- or multi-process (e.g., MPI) setups.
+2
stars, 24h · #12,833
+~10
stars, 7 days · #13,205
+~45
stars, 30 days · #14,878
5.2k
★ total · +0% this week
1.5k
forks · 466 open issues
Star history (as recorded here)
LanguageC++
License—
Created2015-11-14 (10y ago)
Last pushtoday
Watchers164
OwnerOrganization
Seen on GitHub Trending
- 2026-10-23 — #8 daily in c++
GitHub topics
Links
- GitHub: github.com/NVIDIA/nccl
- Homepage: https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/index.html
- Releases · Issues · Hacker News mentions
README
Loading README…
Blurb generated by claude-haiku-5-5 1d ago. Star deltas marked ~ are extrapolated from partial history.