NVIDIA released cuda-oxide: a compiler for Rust code on GPUs

NVIDIA opened the source code for cuda-oxide — an experimental compiler that transforms Rust functions directly into CUDA GPU code. Until now, GPU application developers had to choose between the convenience of Rust and the performance of CUDA.

How It Works

cuda-oxide is a custom backend for the rustc compiler. It takes functions annotated with #[kernel] and runs them through a transformation chain: Rust → Stable MIR → Pliron IR → LLVM IR → PTX (Parallel Thread Execution, machine code for NVIDIA GPU). The main feature: developers write code once. A single `cargo oxide build` command compiles both the CPU part (host) and the GPU part (device). Previously, you had to mix Rust for the host and C/CUDA for the kernels.

Why This Matters

Rust on GPU is the holy grail for developers. CUDA has historically been tied to C/C++. Rust offers memory safety without garbage collection, which is critical for high-performance computing.

• SIMT programming (Single-Instruction Multiple-Thread) becomes more accessible
• Single tooling ecosystem — Cargo works with all targets the same way
• Memory-safety guarantees even for GPU code
• Easier integration of CUDA into existing Rust projects
• Reduces the boundary between host language and kernel language

cuda-oxide v0.1.0 is the first step. The tool is experimental, so production deployments may have limitations and bugs.

What This Means

NVIDIA is investing in the Rust ecosystem for GPU computing. This signals that the world is moving away from mandatory C/C++ for parallel computing. If cuda-oxide reaches a production version, it could become the standard for new CUDA projects in Rust and change the landscape of GPU development.