CUDA.jl

CUDA programming in Julia

The CUDA.jl package is the main programming interface for working with NVIDIA CUDA GPUs using Julia. It features a user-friendly array abstraction, a compiler for writing CUDA kernels in Julia, and wrappers for various CUDA libraries.

Quick start

Before all, make sure you have a recent NVIDIA driver. On Windows, also make sure you have the Visual C++ redistributable installed. You do not need to install the CUDA Toolkit.

CUDA.jl can be installed with the Julia package manager. From the Julia REPL, type ] to enter the Pkg REPL mode and run:

pkg> add CUDA

Or, equivalently, via the Pkg API:

julia> import Pkg; Pkg.add("CUDA")

For an overview of the CUDA toolchain in use, you can run the following command after importing the package:

julia> using CUDA

julia> CUDA.versioninfo()

This may take a while, as it will precompile the package and download a suitable version of the CUDA toolkit. If your GPU is not fully supported, the above command (or any other command that initializes the toolkit) will issue a warning.

For quick testing, you can also use the juliagpu/cuda.jl container image from the GitHub Container Registry, which provides Julia, a precompiled version of CUDA.jl, and a matching CUDA toolkit:

docker run -it --rm --gpus=all ghcr.io/juliagpu/cuda.jl:latest  # other tags available too

For more usage instructions and other information, please refer to the documentation.

Requirements

The latest development version of CUDA.jl requires Julia 1.10 or higher. If you are using an older version of Julia, you need to use a previous version of CUDA.jl. This will happen automatically when you install the package using Julia's package manager.

Note that CUDA.jl may not work with a custom build of Julia; it is recommended that you install Julia using the official binaries or juliaup.

The latest version of CUDA.jl also has certain requirements that cannot be enforced by the package manager:

• Host platform: only 64-bit Linux and Windows are supported;
• Device hardware: only NVIDIA GPUs with compute capability 3.5 (Kepler) or higher are supported;
• NVIDIA driver: a driver for CUDA 11.0 or newer is required;
• CUDA toolkit (in case you need to use your own): only CUDA toolkit 11.4 or newer are supported.

If you cannot meet these requirements, you may need to install an older version of CUDA.jl:

• CUDA.jl v5.3 is the last version with support for PowerPC (removed in v5.4)
• CUDA.jl v4.4 is the last version with support for CUDA 11.0-11.3 (deprecated in v5.0)
• CUDA.jl v4.0 is the last version to work with CUDA 10.2 (removed in v4.1)
• CUDA.jl v3.8 is the last version to work with CUDA 10.1 (removed in v3.9)
• CUDA.jl v1.3 is the last version to work with CUDA 9-10.0 (removed in v2.0)

Supporting and Citing

Much of the software in this ecosystem was developed as part of academic research. If you would like to help support it, please star the repository as such metrics may help us secure funding in the future. If you use our software as part of your research, teaching, or other activities, we would be grateful if you could cite our work. The CITATION.bib file in the root of this repository lists the relevant papers.

Project Status

The package is tested against, and being developed for, Julia 1.10 and above. Main development and testing happens on x86 Linux, but the package is expected to work on Windows and ARM and as well.

Questions and Contributions

Usage questions can be posted on the Julia Discourse forum under the GPU domain and/or in the #gpu channel of the Julia Slack.

Contributions are very welcome, as are feature requests and suggestions. Please open an issue if you encounter any problems.