Infrastructure & deployment
Introducing CUDA Rust: Two Tracks for Writing GPU Kernels
CUDA Rust offers two ways to write native GPU kernels: SIMT for direct thread and memory control, and Tile for compiler-managed work on data tiles.
Why it matters: The choice affects both control and safety boundaries: Tile derives ownership and launch geometry from partitions, while SIMT exposes thread and memory control. The examples show useful checks, not a guarantee of universal memory safety or production readiness.