Registry / data / distributed-ucxx-cu12

distributed-ucxx-cu12

library0.50.1pypypi✓ verified 81d ago

UCX communication module for Dask Distributed, providing high-performance GPU-aware communication using UCX (Unified Communication X). This is the CUDA 12 variant of distributed-ucxx. Version 0.49.0 requires Python >=3.11 and CUDA 12.2+. Released as part of RAPIDS, with monthly release cadence.

data database http-networking gcp workflow

pip install distributed-ucxx-cu12

github ↗package ↗homepage ↗docs ↗

INSTALL

IMPORT

SIG · DISTRIBUTED-UCXX-C

distributed-ucxx-cu12

datapythonv0.50.1

Install

14.8s avg

Import

—

Disk

522MB

Pass rate

5/ 10

Env Coverage5 / 10

glibc

3.9–3.13

musl

3.9–3.13

Install & Compatibility

Where this runs

tested against v0.48.0 · pip install

no network on importno background threads

Install × environment matrix

Each cell = how many times install + import succeeded across repeated harness runs. Partial = flaky.
glibc = Debian/Ubuntu slim · musl = Alpine Linux

musl

py 3.10–3.95 runs

build_error

glibc

py 3.10–3.95 runs

installs and imports cleanly · install 14.8s · import 0.000s · 1006MB

522MB installed

● package 522MB

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

UCXXBackend

✓ from distributed_ucxx import UCXXBackend

✗ from distributed_ucxx import UCX

config

✓ from distributed_ucxx import config

ucxx

✓ from distributed_ucxx import ucxx

Minimal example to start UCX communication using distributed-ucxx-cu12.

from distributed_ucxx import UCX import asyncio async def main(): ucx = UCX() await ucx.start() print("UCX started successfully") await ucx.close() asyncio.run(main())

Debug

Known issues

breakingv0.49.0 replaces numba-cuda runtime dependency with cuda-core. Existing code relying on numba types may break.

fix

Update imports and ensure cuda-core is installed. Remove direct numba-cuda dependency if not explicitly needed.

affects: >=0.49.0

breakingv0.47.0 requires CUDA 12.2+. Older CUDA versions are no longer supported.

fix

Upgrade to CUDA 12.2 or later, or use an older version of distributed-ucxx-cu12 (e.g., 0.46.0).

affects: >=0.47.0

gotchaThe import path is `distributed_ucxx`, not `distributed-ucxx`. Hyphens are not valid in Python module names.

fix

Use `from distributed_ucxx import ...`

affects: all

deprecatedSome older APIs like `ucx_init` are deprecated in favor of `UCX.start()`. Check the documentation for updated patterns.

fix

Use the modern async context manager pattern.

affects: >=0.46.0

Upgrade

Version history

0.50.1latest on PyPI · released Jun 9, 2026

Audit

Dependencies

cuda-corerequiredReplaces numba-cuda runtime dependency starting v0.49.0

ucx-pyrequiredCore UCX Python bindings

distributedrequiredDask Distributed integration

Agent activity

37 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/rapidsai/ucxx ↗

packagehttps://pypi.org/project/distributed-ucxx-cu12/ ↗

homepagehttps://rapids.ai ↗

docshttps://distributed.dask.org/ ↗