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

# libcudf-cu12 — cuDF - GPU Dataframe (C++)

License: permissive · Maintenance: active · Downloads: 469.5K/mo

## What it is and what it does

libcudf-cu12 is the low-level C++ foundation of the RAPIDS GPU data processing suite, providing Apache Arrow-compliant columnar data structures and core algorithms optimized for NVIDIA CUDA 12 GPUs. It handles the heavy lifting for tabular operations—reading, filtering, aggregating, and transforming data—directly on GPU memory, bypassing CPU bottlenecks. The library depends on libkvikio-cu12, librmm-cu12, nvidia-libnvcomp-cu12, nvidia-nvjitlink-cu12, and rapids-logger to function.

The package targets data scientists and engineers working with large tabular datasets who need GPU acceleration. Installation requires a compatible NVIDIA GPU, CUDA 12 runtime, and Linux with specific glibc versions; the wheel distribution is architecture-specific (aarch64 and x86_64). Requires Python 3.11 or later.

Use it for:

- Accelerate large-scale data aggregations and groupby operations on GPU for analytics workflows
- Process multi-gigabyte parquet or CSV files faster than CPU alternatives by keeping data on GPU memory
- Build GPU-accelerated ETL pipelines using libraries built on top of libcudf-cu12's core algorithms
- Integrate GPU tabular processing into Spark jobs via plugins built on libcudf-cu12
- Support GPU-native SQL engines and data processing frameworks that depend on libcudf-cu12

## Worth the install?

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

libcudf-cu12 is a GPU-accelerated C++ library providing Apache Arrow-compliant data structures and fundamental algorithms for tabular data processing on NVIDIA CUDA 12 GPUs.

Yes, if you have an NVIDIA GPU with CUDA 12 and need GPU-accelerated tabular data processing. The library is actively maintained, permissively licensed under Apache-2.0, and has no known vulnerabilities. Install friction is moderate (GPU-specific wheels, glibc version constraints, Python 3.11+), but manageable for teams with GPU infrastructure. Most users will install it indirectly as a dependency rather than directly.

## Install

pip install libcudf-cu12
uv add libcudf-cu12
poetry add libcudf-cu12

## Installing libcudf-cu12

Before you install: Medium install friction due to GPU-specific wheel requirements (manylinux_2_26+ and manylinux_2_27+ for aarch64 and x86_64 respectively). Active maintenance with recent release (8 days old) and strong repository health (9730 stars, last commit 2026-08-14). Requires Python 3.11+.

License in practice: Apache-2.0 permissive license allows commercial and private use with minimal restrictions, making it suitable for production deployments in proprietary systems.

Quickstart:

pip install libcudf-cu12

# libcudf-cu12 is a C++ library with 5 runtime dependencies:
# libkvikio-cu12, librmm-cu12, nvidia-libnvcomp-cu12,
# nvidia-nvjitlink-cu12, rapids-logger

Requires NVIDIA GPU with CUDA 12 support, compatible GPU driver, and Linux system with glibc 2.26+ (aarch64) or 2.27+ (x86_64). Python 3.11 or later.

Verify before relying:

- Whether libcudf-cu12 can be used directly from Python or requires wrapper libraries for practical usage
- Specific GPU compute capability requirements beyond CUDA 12 availability
- Performance characteristics and typical speedup factors vs. CPU-based alternatives for common operations

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: medium
- Maintenance: active
- Downloads: 469.5K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

gpu dataframe library, cuda tabular data processing, gpu accelerated dataframe, arrow gpu implementation, rapids gpu data, nvidia cuda dataframe, gpu columnar storage, gpu-accelerated, rapids, arrow-columnar

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