---
id: dask-cudf-cu12
version: "26.8.0"
license: Apache-2.0
license_treatment: permissive
maintenance: active
---

# dask-cudf-cu12 — Utilities for Dask and cuDF interactions

License: permissive · Maintenance: active · Downloads: 186.9K/mo

## What it is and what it does

Dask cuDF is a GPU-accelerated extension for Dask DataFrame that brings RAPIDS cuDF's pandas-like API to distributed GPU computing. It automatically registers as the 'cudf' backend for Dask, allowing you to write familiar pandas-style code that executes on GPUs instead of CPUs. The package depends on cudf-cu12, cupy-cuda12x, fsspec, numpy, nvidia-ml-py, pandas, and rapids-dask-dependency to provide GPU computation and memory management. It handles coordination between Dask's task scheduler and GPU operations, making it possible to process datasets larger than a single GPU's memory by spilling to host memory.

The package is actively maintained, supports Python 3.11–3.14, and carries no known security vulnerabilities. Single-node multi-GPU workflows are the primary use case. The description notes that multi-node execution requires deploying a distributed cluster separately.

Use it for:

- Process multi-gigabyte Parquet or CSV datasets on a single machine with multiple GPUs faster than CPU-based Dask.
- Run groupby, join, and aggregation operations on GPU-resident data using familiar pandas-style syntax.
- Prototype data pipelines that will later scale to multi-node GPU clusters without rewriting core logic.
- Leverage GPU memory pools and spilling to host memory for workloads that exceed individual GPU VRAM.
- Integrate GPU-accelerated dataframe operations into existing Dask workflows.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Dask cuDF extends Dask DataFrame with a GPU-accelerated backend, enabling parallel and larger-than-memory data processing on NVIDIA GPUs using a pandas-like API.

Yes, if you have NVIDIA GPUs and need to process large datasets faster than CPU Dask. The low install friction, active maintenance, permissive Apache-2.0 license, and zero known vulnerabilities make it a solid choice. Requires CUDA 12 runtime and GPU hardware; not suitable for CPU-only environments.

## Install

pip install dask-cudf-cu12
uv add dask-cudf-cu12
poetry add dask-cudf-cu12

## Installing dask-cudf-cu12

Before you install: Low install friction; pure Python wheel. Active maintenance with recent releases. Requires CUDA 12 runtime and GPU libraries (cudf-cu12, cupy-cuda12x, nvidia-ml-py) as dependencies, which may require system-level NVIDIA driver setup.

License in practice: Apache-2.0 permissive license allows commercial and private use with minimal restrictions.

Quickstart:

pip install dask-cudf-cu12

import dask.dataframe as dd
from cupy import cuda

df = dd.read_parquet("/path/to/data/")
result = df.groupby('item')['price'].mean().compute()

Requires NVIDIA GPU(s) with CUDA 12 support and compatible drivers; cudf-cu12 and cupy-cuda12x must be installed. Python >=3.11 required.

Verify before relying:

- Whether multi-GPU scaling on a single node works without additional cluster deployment tooling.
- Performance characteristics and memory overhead compared to single-GPU cuDF or CPU Dask DataFrame.
- Compatibility with specific NVIDIA GPU architectures and driver versions beyond CUDA 12.

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 186.9K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

gpu dataframe processing, dask cuda dataframe, parallel gpu data analysis, rapids dask integration, gpu-accelerated pandas, distributed gpu computing, dask cudf backend, gpu-computing, distributed-dataframes, rapids

[View on SkillFed](https://skillfed.io/packages/dask-cudf-cu12) · [View on PyPI](https://pypi.org/project/dask-cudf-cu12/)