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

# pylibcudf-cu12 — pylibcudf - Python bindings for libcudf

License: permissive · Maintenance: active · Downloads: 411.0K/mo

## What it is and what it does

pylibcudf-cu12 is a low-level Python binding layer for libcudf, the core CUDA C++ library in RAPIDS that implements GPU-accelerated tabular data operations. It exposes libcudf's Arrow-compliant data structures and algorithms to Python via Cython, serving as the foundation for higher-level RAPIDS libraries like cudf and dask-cudf.

This package is designed for developers building GPU-accelerated data processing applications. It requires a CUDA 12 environment and compatible NVIDIA GPU hardware. The package has four runtime dependencies: libcudf-cu12 (the C++ library itself), cuda-bindings, rmm-cu12 (RAPIDS memory manager), and nvtx (NVIDIA tracing tools). Most users interact with pylibcudf indirectly through cudf's pandas-like API or cudf.pandas, which provides a zero-code-change accelerator for existing pandas code.

Use it for:

- Building custom GPU-accelerated data processing pipelines that need direct access to libcudf algorithms.
- Integrating GPU tabular operations into Spark via Spark RAPIDS or other distributed computing frameworks.
- Developing specialized data transformations where cudf's high-level API is insufficient.
- Creating GPU-native SQL engines or query processors that execute on tabular data.
- Accelerating ETL workflows for large datasets on systems with NVIDIA GPUs and CUDA 12.

## Worth the install?

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

pylibcudf-cu12 provides Python bindings for libcudf, a CUDA C++ library that accelerates tabular data operations on NVIDIA GPUs using Apache Arrow data structures.

Yes, if you have an NVIDIA GPU with CUDA 12 and need GPU-accelerated tabular data processing. Install it as a dependency of cudf or higher-level RAPIDS libraries rather than directly unless you need low-level libcudf access. Active maintenance, no known vulnerabilities, and permissive Apache 2.0 license make it production-ready. Medium install friction is expected given the specialized CUDA runtime requirements.

## Install

pip install pylibcudf-cu12
uv add pylibcudf-cu12
poetry add pylibcudf-cu12

## Installing pylibcudf-cu12

Before you install: Medium install friction due to CUDA 12 runtime dependencies (libcudf-cu12, cuda-bindings, rmm-cu12, nvtx). Requires matching your CUDA version to the cu12 suffix. Active maintenance with recent releases; repository is well-maintained with 9730 stars.

License in practice: Apache 2.0 permissive license allows commercial and private use with minimal restrictions, making it suitable for most production environments.

Quickstart:

pip install pylibcudf-cu12

import pylibcudf
# Access libcudf functionality through Cython bindings
# Typically used indirectly via cudf or other RAPIDS libraries

Requires CUDA 12 runtime environment and compatible NVIDIA GPU; CUDA version must match the cu12 suffix in the package name.

Verify before relying:

- Whether pylibcudf-cu12 is typically used directly or primarily as a dependency of higher-level RAPIDS libraries (cudf, dask-cudf).
- Performance characteristics and typical speedup factors compared to CPU-based alternatives for specific workloads.

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: medium
- Maintenance: active
- Downloads: 411.0K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

gpu dataframe processing, cuda accelerated data operations, gpu tabular data library, rapids libcudf python bindings, arrow gpu acceleration, nvidia gpu data processing, cuda dataframe operations, gpu-acceleration, cuda, rapids

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