Nvidia introduces CUDA Rust with two paths for writing native GPU kernels in Rust across SIMT and tile-based models

Nvidia announced CUDA Rust, a push to support native GPU programming in Rust alongside CUDA C++ and CUDA Python. The effort includes cuda-oxide, an early-alpha compiler path for SIMT-style kernels, and cutile-rs, a stable Rust tile-based programming model already used in some inference projects.

Rust support could make GPU programming safer and more attractive to developers building AI and high-performance computing software. By offering both low-level and higher-level approaches, Nvidia is signaling that memory safety and language flexibility are becoming more important in accelerator software ecosystems.