---
id: jax-cuda12-plugin
version: "0.11.0"
license: Apache-2.0
license_treatment: permissive
maintenance: active
---

# jax-cuda12-plugin — JAX Plugin for NVIDIA GPUs

License: permissive · Maintenance: active · Downloads: 782.4K/mo

## What it is and what it does

jax-cuda12-plugin is a runtime plugin that bridges JAX (a numerical computing library for automatic differentiation and XLA compilation) to NVIDIA GPUs via CUDA 12. When installed, it registers GPU support as a backend for JAX, allowing numerical operations and machine learning computations to execute on NVIDIA hardware instead of CPU.

The plugin depends on jax-cuda12-pjrt, a compiled component that handles low-level GPU communication. It is distributed as platform-specific wheels for Linux x86_64 and aarch64 only, and requires Python 3.12 or later. Installation is straightforward via pip, but GPU availability and proper CUDA environment setup are prerequisites for runtime use.

Use it for:

- Accelerate machine learning training by offloading gradient computation and matrix operations to NVIDIA GPUs.
- Speed up numerical simulations and scientific computing workloads using JAX's automatic differentiation and JIT compilation.
- Enable per-example gradient computation and vectorized operations on GPU for batch processing in deep learning pipelines.
- Scale JAX programs across multiple GPUs using data parallelism and sharding strategies on NVIDIA hardware.

## Worth the install?

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

Enables JAX to run numerical computations and machine learning workloads on NVIDIA GPUs with CUDA 12 support.

Yes, if you have an NVIDIA GPU with CUDA 12 support and want to use JAX for accelerated numerical computing or machine learning. The plugin is actively maintained, has no known vulnerabilities, and carries a permissive license. Install friction is moderate due to platform specificity and compiled dependencies, but this is typical for GPU acceleration packages. Not applicable on systems without NVIDIA GPUs or on unsupported platforms.

## Install

pip install jax-cuda12-plugin
uv add jax-cuda12-plugin
poetry add jax-cuda12-plugin

## Installing jax-cuda12-plugin

Before you install: Medium install friction due to platform-specific wheels for Linux x86_64 and aarch64 only, plus a compiled runtime dependency on jax-cuda12-pjrt. Actively maintained with recent releases and high repository activity (36158 stars).

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

Quickstart:

pip install jax-cuda12-plugin
import jax
import jax.numpy as jnp
# GPU acceleration is automatically available for JAX operations
result = jax.jit(lambda x: jnp.dot(x, x))(jnp.ones((100, 100)))

Requires NVIDIA GPU with CUDA 12 support and Linux x86_64 or aarch64 platform; Python 3.12 or later.

Verify before relying:

- Whether CUDA 12 runtime libraries must be pre-installed on the system for the plugin to function.
- Performance characteristics and memory overhead compared to CPU-only JAX execution.
- Compatibility with specific NVIDIA GPU architectures and CUDA driver versions.

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: medium
- Maintenance: active
- Downloads: 782.4K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

jax nvidia gpu acceleration, cuda 12 gpu plugin, jax gpu backend, nvidia cuda jax support, gpu-accelerated numerical computing, gpu-acceleration, cuda, numerical-computing

[View on SkillFed](https://skillfed.io/packages/jax-cuda12-plugin) · [View on PyPI](https://pypi.org/project/jax-cuda12-plugin/)