Installing JAX unlocks high-performance numerical computing and machine learning research on your machine. This guide walks you through what JAX is, why it matters, and how to set it up correctly across different environments.
Whether you are running experiments on CPU, GPU, or TPU, a clean JAX installation reduces debugging time and keeps your workflows reproducible. The steps below focus on stable, production-friendly approaches.
| Component | Version | Purpose | Notes |
|---|---|---|---|
| Python | 3.9–3.11 | Runtime language | Avoid 3.13 for mainstream JAX use |
| JAX | Latest stable release | Core library | Includes CPU, GPU, and TPU support |
| GPU Driver | Rounded driver version | Hardware interface | Matches CUDA and cuDNN requirements |
| CUDA | Matched toolkit | GPU compute | Required for Nvidia GPU acceleration |
| cuDNN | Aligned build | Deep learning primitives | Essential for neural network performance |
Prerequisites and System Preparation
Before you install JAX, verify your hardware and operating system compatibility. JAX supports major Linux distributions, recent macOS versions, and Windows through WSL2 for GPU workloads.
Check your Python version, available RAM, and GPU specs if applicable. For GPU or TPU support, ensure you have the required driver stack and compatible versions of CUDA and cuDNN installed.
Create a virtual environment to keep dependencies isolated. Using venv or conda prevents version clashes with other projects and simplifies future upgrades.
Install JAX on CPU for Quick Experiments
CPU installation is the simplest way to get started with JAX. It requires no external GPU drivers and runs immediately on most machines.
Use pip to pull the CPU-only package from PyPI. This variant is ideal for prototyping, debugging, and lightweight workloads without GPU acceleration.
After installation, run a quick sanity check by importing JAX and printing local device count to confirm the CPU backend is detected.
Install JAX with GPU Support on Linux
GPU acceleration dramatically speeds up large matrix operations and neural network training. On Linux, you must align JAX, CUDA, and cuDNN versions carefully.
Install the GPU-enabled JAX package and verify that CUDA libraries are discoverable. Use nvidia-smi and nvcc checks to confirm driver and toolkit compatibility before running complex models.
Keep your graphics driver and CUDA toolkit updated in a coordinated manner to avoid subtle runtime errors or performance regressions.
Install JAX on Windows and macOS
On Windows, use WSL2 to run a Linux environment for GPU workloads, because native CUDA support is limited. Inside WSL2, follow the Linux GPU installation steps for a consistent experience.
macOS users can install the CPU version via pip and rely on Apple’s Metal backend for some acceleration where available. Avoid mismatched Python versions and system security settings that may block pip installs.
Test each installation by importing jax and checking jax.devices() to see which accelerators are visible to the runtime.
Optimized Setup and Maintenance
Regular maintenance keeps JAX performant and avoids broken workflows due to version drift. Stick with well-tested version combinations and document your setup for team reproducibility.
- Use a virtual environment or conda env for isolation
- Match CUDA and cuDNN versions to the JAX release notes
- Pin package versions in requirements.txt for reproducibility
- Periodically test a minimal script to verify device detection
- Back up working configurations before major upgrades
FAQ
Reader questions
Why does JAX say GPU not found after pip install?
Confirm that CUDA and cuDNN versions match the JAX build, your NVIDIA driver supports the CUDA toolkit, and the PATH and LD_LIBRARY_PATH include the correct library locations.
Can I install JAX in a conda environment?
Yes, create a conda environment and use pip inside it, or use conda-forge channels for CPU builds. Maintain consistency between Python version and package channels to avoid conflicts.
How do I verify my installation is using the GPU?
Run Python code that prints jax.devices() and checks for gpu or cuda entries. You can also execute a simple jit-compiled function and monitor GPU utilization with nvidia-smi.
What should I do if tests fail after upgrading JAX?
Reinstall CUDA and cuDNN to match the new JAX version, clear pip cache, and recreate your virtual environment. Consult the JAX release notes for known compatibility changes.