App
nvidia / nccl
Optimized primitives for collective multi-GPU communication
This connects Repowise to nvidia/nccl's code. It does not install nccl.
Repowise maps public repos from their source. Not affiliated with nvidia. What is Repowise?
github.com/nvidia/ncclIndexed at 9287574
What is nccl?
NCCL is a stand-alone library of standard communication routines for GPUs, including all-reduce, all-gather, reduce, broadcast and reduce-scatter. It runs these operations across any number of GPUs installed in one node or spread over several nodes. The project is written mainly in C++ and includes Python bindings. To install it, run git clone https://github.com/nvidia/nccl.git and follow the steps below. Its code health is 5.7 out of 10 (Fair), ranked #229 of 4,000+ indexed repos. The last commit to nccl was on Jun 10, 2026.
As of commit 9287574, indexed Jul 14, 2026. What it is comes from a model reading the README; the rest is measured from the source and git history, not from the maintainer.
For agentsnccl.md
Repo facts: type, license, health, rank, contributors
| Type | App |
|---|---|
| Code health | 5.7 / 10 (Fair) |
| Health rank | #229 of 4,000+ indexed repos |
| Last commit | Jun 10, 2026 |
| Bus factor | 2 |
| Size | 834 files, 172,648 lines |
Install nccl
Commands copied from the README
- 1Clone the nccl repository.
git clone https://github.com/nvidia/nccl.git - 2Change into the nccl directory.
cd nccl - 3Build the library into the build directory.
make -j src.build - 4Build a Debian package on Debian or Ubuntu systems.
make pkg.debian.build
Releases on GitHub (latest v2.32.6-1, Nov 9, 2026)
Explore nccl
- DocumentationModule-by-module documentation, written from the source.
- ArchitectureDependency graph, layers and third-party dependencies.
- Code healthEvery file's score and the findings behind it.
- CommitsEach commit's effect on code health.
- OverviewEverything else the index found: contributors, decisions, what changed this week.
Ask about this code in your agent
This connects Repowise to nccl's code. It does not install nccl. To install nccl, use the install steps.
Useful if you want to change or fork nccl: your agent can ask how it is put together and get answers that cite its files.
Ask about nccl
Answers come from nccl's README and source, with the files they cite. Two free answers a day without an account.
Want this for your own codebase? Map your repo free
About this page
Get a page like this for your own codebase
Repowise built this map of nvidia/nccl from its source and git history. It parses every symbol, draws the dependency graph, scores each file's health from complexity, duplication, coverage and churn, finds hotspots and owners in the history, and writes documentation for every module. Your coding agent reads the same index over MCP, so its answers cite real files.
Free for 2 public repos, with a page like this one. No card. See how Repowise works
Prefer local? pip install repowise && repowise init indexes your own checkout, no account needed. Star Repowise on GitHub
nvidia/nccl FAQ
What is NCCL?
NCCL is a stand-alone library of standard communication routines for GPUs. It implements operations such as all-reduce, all-gather, reduce, broadcast and reduce-scatter across multiple GPUs.
How does NCCL work?
NCCL runs collective communication between GPUs in a node or across several nodes, and it can be called from single-process or multi-process applications such as MPI programs. Its source is built into the library with make src.build.
How do I install NCCL?
NCCL is built from source with make -j src.build in the nccl directory, and installable packages are made with targets such as make pkg.debian.build or make pkg.txz.build. A custom CUDA location can be set with CUDA_HOME. See the install steps
What language is NCCL written in?
NCCL is written mainly in C++, with Python bindings in the nccl4py package and a Makefile-based build.
How is the NCCL code organized?
The core library is in the src directory, extra components are in contrib, and the bindings directory holds the Python package. Example programs are kept in docs/examples.
When should I use NCCL?
NCCL is for applications that run on multiple GPUs and need communication between them, whether in one process or in several processes such as MPI jobs. It supports GPUs in a single node or across multiple nodes.
Code health
Hotspots average 2.9, the place fixes pay off most.
On the leaderboards:#9 of 12 C++ repos
Key features of nccl
- Implements all-reduce, all-gather, reduce, broadcast and reduce-scatter collective operations, as well as any send and receive based communication pattern.
- Supports communication between GPUs installed in a single node or across multiple nodes.
- Works over PCIe, NVLink, NVswitch, InfiniBand Verbs and TCP/IP sockets.
- Can be used in single-process or multi-process applications, such as those using MPI.
How nvidia/nccl works
The library source lives in src, the largest module with 404 files, and the Makefile target src.build compiles it into build/. Other Makefile targets such as pkg.debian.build, pkg.redhat.build, pkg.txz.build and pkg.python_wheel.build package the built library under build/pkg. The contrib directory holds 153 files of additional components, including contrib/nccl_ep/nccl_ep.cc, which is one of the listed entry points. The bindings directory has 46 files, including the nccl4py Python package with examples under bindings/nccl4py/examples. The docs directory has 95 files, including the C example docs/examples/01_communicators/01_multiple_devices_single_process/c/main.cc, and src/dev_runtime.cc and src/graph/topo.cc are listed among the hotspot files.
The codebase has 834 files and 13,104 symbols in 8 modules, led by C++, Python and Makefile. Code health is 5.6 out of 10, rated fair.
Maintain nvidia/nccl?
Claim this page, add a badge to the README, or re-index it.
Claim this page
Sign in with GitHub and we check that you are an admin or maintainer of nvidia/nccl. Then add a badge to the README that links readers here.
Add a badge to your README
Readers click through to this page and can ask nccl's code a question. A badge also re-indexes the repo every week and marks you as the verified maintainer.
[](https://repowise.dev/repo/nvidia/nccl?src=badge_ask)Paste it at the top of README.md, under the title.
Added it already? Verify to get the mark now instead of tomorrow.
Need HTML or the larger health card? More formatsHide more formats
This badge as HTML
<a href="https://repowise.dev/repo/nvidia/nccl?src=badge_ask"><img src="https://api.repowise.dev/badge/wiki/nvidia/nccl.svg?label=ask" alt="repowise: Ask this codebase"></a>Code health card
The score, its change since the last index, and when it was measured. Follows the reader's light or dark theme.
<a href="https://repowise.dev/repo/nvidia/nccl?src=badge_card"><picture><source media="(prefers-color-scheme: light)" srcset="https://api.repowise.dev/badge/health/nvidia/nccl.svg?style=card&theme=light"><img src="https://api.repowise.dev/badge/health/nvidia/nccl.svg?style=card" alt="repowise code health" width="500" height="168"></picture></a>