NVIDIA CUDA Rust
NVIDIA CUDA™ Rust enables developers to natively build accelerated applications in Rust. Developers can get started writing CUDA SIMT and CUDA Tile kernels via cuda-oxide and cutile-rs, respectively. cuda-oxide compiles pure Rust kernels to PTX with a single source, and cutile-rs brings tile-based kernel authoring and a new level of safety to GPU programming.
Demo
Pure Rust GEMM on NVIDIA Blackwell With cuda-oxide
See cuda-oxide compile and run a matrix-multiply (GEMM) kernel written as a plain Rust function on NVIDIA Blackwell hardware. No CUDA C++, no DSL, no foreign-language kernel source.
This walkthrough shows what's possible with cuda-oxide today, tracing the path from Rust source through PTX to real GPU execution and examining what performance headroom remains as the project matures.
Why CUDA Rust
The systems layer of AI is increasingly written in Rust: inference engines, serving infrastructure, drivers, and agent runtimes. The GPU kernel has been the exception, still authored in another language. NVIDIA CUDA Rust is closing that gap.
Two experimental projects are public today. cuda-oxide compiles pure Rust kernels to PTX with no DSL or foreign-language kernel source. cutile-rs brings tile-based kernel authoring with Rust ownership across host, launch boundary, and device code. Both are safe by default, with local escape hatches where the work requires them, and both reject racing code at compile time rather than letting it fail unpredictably later.
It's early, it's open, and what you build now will shape what comes next.
Get Started With CUDA Rust
Both projects are openly available on GitHub. Clone the repository, read the README, and file issues to help shape what comes next.
cutile-rs repository
cuTile-rs provides an implementation of the CUDA Tile programming model.
cuda-oxide repository
cuda-oxide provides a single-source CUDA SIMT implementation using standard RUST tooling.
More Resources
Ethical AI
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI Concerns here.