# Rust-to-CUDA

> **TL;DR.** Rust-to-CUDA describes compiling standard Rust code directly into PTX — NVIDIA's GPU assembly — enabling developers to write GPU kernels without C++, domain-specific languages, or foreign function interfaces.

- **Category:** Developer Tools / GPU / Systems Programming
- **Stage:** validating
- **Age:** 36 days
- **Origin date:** 2026-05-09
- **First detected:** 2026-05-09
- **Canonical URL:** https://earlyterms.com/term/rust-to-cuda
- **Sources:** 7 primary URLs

## Definition

Rust-to-CUDA describes compiling standard Rust code directly into PTX — NVIDIA's GPU assembly — enabling developers to write GPU kernels without C++, domain-specific languages, or foreign function interfaces. It is the practice that NVIDIA's own [cuda-oxide](https://github.com/NVlabs/cuda-oxide) compiler makes possible.

The term crystallized on May 9, 2026 when NVIDIA Labs released cuda-oxide v0.1.0 under NVlabs: a custom rustc codegen backend that transforms Rust through MIR → Pliron IR → LLVM IR → PTX, with support for generics, closures, async composition, and Hopper/Blackwell intrinsics — the first official NVIDIA-backed path for writing GPU kernels in Rust.

## Example

A cuda-oxide kernel for matrix multiplication achieved 868 TFLOPS on B200 hardware — roughly 58% of cuBLAS — written in plain Rust with no C++ and no DSL. The project's 60+ examples range from a basic thread-indexing hello-world to tile-based GEMM and warp-level reductions, all compiled with cargo.

## Analogy

Think of it as replacing the C++ torch kernels under the hood with Rust, without rewriting the calling API.

## Why it's emerging now

NVIDIA Labs shipped cuda-oxide v0.1.0 on May 9, 2026 — its first official, org-backed Rust-to-PTX compiler. This ends five years of community-only attempts (the Rust-CUDA project spent 2021-2024 dormant, rebooted in Jan 2025 with no NVIDIA involvement) and signals that the world's dominant GPU vendor now considers Rust a first-class language for GPU kernel authoring.

## Related terms

- *child:* cuda-oxide
- *competitor:* Rust-GPU
- *competitor:* rustc_codegen_nvvm
- *competitor:* CubeCL
- *related:* PTX
- *parent:* CUDA
- *competitor:* Mojo
- *competitor:* Slang
- *related:* cudarc
- *related:* Pliron
- *related:* bun-s-rust
- *related:* nemotron-ultra

## Sources

1. [cuda-oxide GitHub — NVlabs official repo](https://github.com/NVlabs/cuda-oxide)
2. [cuda-oxide Book — official documentation](https://nvlabs.github.io/cuda-oxide/index.html)
3. [HN: CUDA-oxide: Nvidia's official Rust to CUDA compiler (424 pts)](https://news.ycombinator.com/item?id=48096692)
4. [MarkTechPost: NVIDIA AI releases cuda-oxide — Rust-to-PTX compiler backend](https://www.marktechpost.com/2026/05/09/nvidia-ai-just-released-cuda-oxide-an-experimental-rust-to-cuda-compiler-backend-that-compiles-simt-gpu-kernels-directly-to-ptx/)
5. [ByteIota: NVIDIA CUDA-Oxide — Official Rust-to-CUDA Compiler](https://byteiota.com/nvidia-cuda-oxide-official-rust-to-cuda-compiler-released/)
6. [Rust+GPU Ecosystem — where cuda-oxide fits vs rust-cuda, wgpu, CubeCL](https://nvlabs.github.io/cuda-oxide/appendix/ecosystem.html)
7. [Rust-GPU Blog: Rust-CUDA Project Rebooted (community context)](https://rust-gpu.github.io/blog/2025/03/18/rust-cuda-update/)

---
_Generated by EarlyTerms · https://earlyterms.com/term/rust-to-cuda_