NVlabs/cuda-oxide

Health v2 · maintenance only

Flags: young

Adoption not part of the score

3097 stars · 243 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

cuda-oxide is an experimental rustc codegen backend that compiles Rust GPU kernels directly to CUDA PTX, letting you write SIMT device code in idiomatic Rust with no DSLs or foreign bindings. It includes device-side abstractions (shared memory, atomics, warp ops, TMA) and a host-side runtime for memory management and kernel launching.

Use cases

• write cuda kernels in rust
• compile rust to ptx
• gpu programming without c++
• single-source host and device code in rust
• launch gpu kernels from rust
• use tensor memory accelerator from rust
• async gpu programming in rust

When to choose

• you want to write GPU kernels in pure idiomatic Rust without DSLs or C++ bindings
• you want host and device code in the same Rust source compiled with cargo
• you need device-side features like shared memory, warp ops, clusters, or TMA from Rust
• you're exploring the Rust GPU ecosystem and can tolerate alpha-quality software

When to avoid

• you need a stable, production-ready CUDA toolchain — the project is an early alpha with API breakage
• you need full CUDA feature coverage or mature tooling comparable to CUDA C++/nvcc
• you don't use NVIDIA GPUs or Rust
• you need guaranteed correctness for safety-critical workloads given the experimental safety model

Facets

library · maturity experimental

compiler gpu-computing sdk gpu-computing programming-languages compilers developer-tools rust cuda ptx rustc-backend simt kernels nvidia heterogeneous-computing no-dsl linux nvidia-gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVlabs/cuda-oxide")