skillfed

oneccl

Intel® oneAPI Collective Communications Library Runtime Environment

oneccl v2022.1.1 148.1K downloads/30d#11,045 on PyPI
License unclear Intel Simplified Software License Active released

What it is and what it does

oneccl is Intel's oneAPI Collective Communications Library runtime, a low-level component for distributed machine learning training. It provides optimized communication primitives (allreduce, broadcast, etc.) across multiple compute nodes, typically used as a backend by higher-level distributed training frameworks. The package itself is a runtime environment rather than a user-facing API—it is usually installed as a dependency of other tools rather than used directly.

The package carries medium install friction because it depends on two Intel-specific runtime libraries (impi-rt and intel-sycl-rt) and is distributed only as a pre-built wheel for x86_64 Linux. It is actively maintained and carries no known security vulnerabilities, but its license is proprietary and not SPDX-identified, which may matter for compliance-sensitive deployments.

Use it for:

  • Backend runtime for distributed PyTorch or TensorFlow training on Intel hardware clusters.
  • Enabling multi-node collective operations in custom distributed training scripts.
  • Optimizing communication latency in large-scale machine learning workloads on Intel systems.

Worth the install?

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

Provides Intel's oneAPI Collective Communications Library runtime for distributed training workloads, enabling scalable communication across multiple compute nodes.

Yes, if you are running distributed training on Intel hardware and your training framework (PyTorch, TensorFlow, etc.) lists oneccl as a dependency or recommends it for performance. No, if you are not doing distributed training or are using non-Intel hardware. Clarify the proprietary license terms with Intel before production use in commercial or regulated environments.

Install

oneccl on PyPI

pip

pip install oneccl

uv

uv add oneccl

poetry

poetry add oneccl

Installing oneccl

Before you install

Medium install friction due to platform-specific wheels and two Intel runtime dependencies (impi-rt and intel-sycl-rt). Package is actively maintained with recent releases.

License in practice

Licensed under Intel Simplified Software License (proprietary, not SPDX-identified). License treatment is unclear—review Intel's terms before deploying in production or commercial contexts.

Quickstart

pip install oneccl
import oneccl
# Runtime depends on impi-rt and intel-sycl-rt being available in the environment

Requires impi-rt and intel-sycl-rt runtime dependencies; x86_64 Linux only (manylinux_2_28).

Verify before relying

  • Whether Python version support is truly unspecified or if there are undocumented constraints.
  • Whether the two runtime dependencies (impi-rt, intel-sycl-rt) are automatically installed or require manual setup.
  • Compatibility with non-x86_64 architectures or non-Linux operating systems.

Package facts

License Intel Simplified Software License (unclear)
Python support not specified
Install friction medium — platform-specific wheel
Runtime dependencies 2 — impi-rt, intel-sycl-rt
Maintenance actively maintained — 30 days since the last release
First released
Downloads 148,063/month — #11,045 on PyPI (30-day window, as of 2026-08-14)
Known vulnerabilities none known (OSV.dev, checked 2026-08-14)

Evidence: oneccl-2022.1.1-py2.py3-none-manylinux_2_28_x86_64.whl

Development Status :: 5 - Production/StableIntended Audience :: DevelopersIntended Audience :: EducationIntended Audience :: Other AudienceIntended Audience :: Science/ResearchIntended Audience :: System AdministratorsLicense :: Other/Proprietary LicenseOperating System :: POSIX :: LinuxTopic :: Software Development :: Libraries

Tags

distributed training communicationcollective communications libraryintel oneapi runtimemulti-node trainingscalable distributed computing
distributed-trainingintel-hardware

More Libraries packages

Further reading