nvidia-nccl-cu11

What it is and what it does

This package bundles NVIDIA's Collective Communication Library (NCCL) runtime for CUDA 11, implementing collective communication routines like all-reduce, all-gather, reduce, broadcast, and reduce-scatter. It has been optimized to achieve high bandwidth on any platform using PCIe, NVLink, NVswitch, as well as networking using InfiniBand Verbs or TCP/IP sockets. The package is not a standalone library you call directly from Python; instead, it serves as a runtime dependency for frameworks that need to coordinate computation across multiple GPUs.

Because this is a platform-specific wheel (manylinux2014 x86_64 only) and the package has been dormant for 863 days with no recent maintenance, it is primarily useful for legacy projects already pinned to CUDA 11. New projects should evaluate whether a current CUDA version and its corresponding NCCL package better suit their needs.

Use it for:

• Provide the NCCL runtime for distributed deep learning frameworks on CUDA 11 systems.
• Support legacy production deployments that are locked to CUDA 11 and need the matching NCCL library.
• Enable multi-GPU collective communication on systems using PCIe, NVLink, NVswitch, or InfiniBand.
• Facilitate research or development on multi-GPU systems where collective communication is critical.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides NVIDIA's collective communication library (NCCL) runtime for GPU-accelerated all-reduce, all-gather, reduce, broadcast, and reduce-scatter operations optimized for CUDA 11.

Yes, but only if you are locked to CUDA 11 and require NCCL for multi-GPU work. The package is dormant and will not receive updates, so it is suitable only for stable, legacy deployments. If you are starting a new project, prefer a current CUDA version and its corresponding NCCL package instead. Verify NVIDIA's licensing terms before production use.

Install

nvidia-nccl-cu11 on PyPI

pip install nvidia-nccl-cu11

uv add nvidia-nccl-cu11

poetry add nvidia-nccl-cu11

Installing nvidia-nccl-cu11

Before you install

Medium install friction due to platform-specific wheel (manylinux2014 x86_64). Package is dormant—last release was 863 days ago—so expect no active maintenance or bug fixes going forward.

License in practice

Licensed under NVIDIA Proprietary Software with unclear treatment. Review NVIDIA's licensing terms before deploying in production or redistributing.

Quickstart

pip install nvidia-nccl-cu11==2.21.5
# Typically imported indirectly through frameworks that depend on this runtime

Requires NVIDIA CUDA 11 and compatible GPU hardware; x86_64 Linux only (manylinux2014 wheel). Not a standalone Python library—used as a runtime dependency.

Verify before relying

• Whether this package is still the recommended way to install NCCL for CUDA 11, or if newer CUDA versions have superseded it.
• Compatibility with specific framework versions that depend on this NCCL runtime.
• How to invoke NCCL primitives directly from Python code using this package.

Package facts

Evidence: nvidia_nccl_cu11-2.21.5-py3-none-manylinux2014_x86_64.whl

Keywords: cuda, nvidia, runtime, machine learning, deep learning

Tags

More Software Development packages

Further reading