GitHub - warpfront/hipfire: RDNA-native LLM inference engine in Rust.
RDNA-native LLM inference engine in Rust. . Contribute to warpfront/hipfire development by creating an account on GitHub.
Welcome! Type "help" for available commands.
$
Summary
NVIDIA introduced CUDA Rust in September 2026, enabling GPU kernels to be written in Rust and compiled natively to PTX.
Two programming tracks are available: SIMT via the cuda-oxide rustc codegen backend, and Tile via the Tile IR compiler. The SIMT track requires Linux, compute capability 8.0+, CUDA 12.x, and a pinned nightly toolchain, and enforces memory safety through DisjointSlice types and validated launch contracts.
