raft-dask-cu12

What it is and what it does

raft-dask-cu12 is the Dask integration layer for RAFT, enabling distributed GPU-accelerated computing across multiple nodes and GPUs. It wraps RAFT's host-accessible runtime APIs—which don't require a CUDA compiler—and layers them atop Dask's distributed task scheduler and NVIDIA's NCCL for inter-GPU communication. The package is part of the RAPIDS ecosystem and targets application developers and data source providers building high-performance distributed algorithms, not data scientists doing exploratory work.

The package depends on dask-cuda for GPU task scheduling, distributed-ucxx-cu12 for optimized communication, libraft-cu12 for the core accelerated primitives, and pylibraft-cu12 for Python bindings. It is actively maintained, supports Python 3.11 through 3.14, and carries no known security vulnerabilities. Installation requires a CUDA 12-compatible GPU environment and brings medium friction due to its compiled dependencies.

Use it for:

• Build distributed machine learning algorithms that scale across GPU clusters using RAFT's low-level primitives as building blocks.
• Accelerate multi-node data processing workflows by offloading linear algebra, sparse operations, and statistical computations to GPUs.
• Integrate GPU-accelerated RAFT functions into existing Dask applications without rewriting core algorithms.
• Develop high-performance solvers and optimization routines that benefit from RAFT's centralized, reusable GPU kernels.

Worth the install?

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

Provides multi-node multi-GPU distributed computing infrastructure for building GPU-accelerated algorithms with Dask, wrapping RAFT's host-accessible runtime APIs.

Yes, if you are building distributed GPU-accelerated applications on multi-node clusters and want to leverage RAFT's low-level primitives through Dask. The package is actively maintained, carries no known vulnerabilities, and is permissively licensed. Install friction is moderate due to CUDA dependencies, but this is expected for GPU-targeting libraries. Not recommended for single-node work or exploratory data science—use pylibraft-cu12 directly or RAPIDS cuML instead.

Install

raft-dask-cu12 on PyPI

pip install raft-dask-cu12

uv add raft-dask-cu12

poetry add raft-dask-cu12

Installing raft-dask-cu12

Before you install

Medium install friction due to 6 runtime dependencies including CUDA-specific packages (libraft-cu12, nvidia-nccl-cu12, pylibraft-cu12) and Dask infrastructure. Actively maintained with recent releases; requires Python 3.11+.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions, typical for RAPIDS ecosystem components.

Quickstart

pip install raft-dask-cu12

from raft_dask_cu12 import ...
# Use with dask.distributed and GPU-accelerated RAFT primitives

Requires NVIDIA GPU with CUDA support and compatible CUDA 12 environment; intended for multi-node GPU clusters, not single-CPU systems.

Verify before relying

• Specific distributed algorithms and primitives exposed through raft-dask-cu12 Python API beyond the general RAFT categories mentioned.
• Whether this package is the recommended entry point for Dask users or if pylibraft-cu12 is preferred for single-node GPU work.
• Performance characteristics and scalability limits for typical multi-node multi-GPU deployments.

Package facts

Evidence: raft_dask_cu12-26.8.0-cp311-abi3-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; raft_dask_cu12-26.8.0-cp311-abi3-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl

Tags

More Distributed Computing packages

Further reading