Introducing CUDA Rust: Two Tracks for Writing GPU Kernels

NVIDIA's CUDA Rust work offers two ways to write the code that runs on a GPU. With cuda-oxide, developers specify individual thread behavior and retain control over threads and memory. With cutile-rs, they describe operations on blocks of data, and the compiler handles their mapping onto GPU threads. NVIDIA recommends starting with the tile approach and using the lower-level option when that control is needed. The announcement also states that both projects are early-stage and neither is production-ready, positioning them as tools to experiment with as native Rust GPU support develops.