XNews

NVIDIA introduces CUDA Tile in CUDA 13.1 to shift programming from threads to tiles

NVIDIA introduces CUDA Tile in CUDA 13.1 to shift programming from threads to tiles

CUDA 13.1 introduces CUDA Tile, a tile-based programming model that shifts developer focus from managing thousands of threads (SIMT) to operating on data blocks (tiles). CUDA Tile abstracts complex hardware (Tensor Cores, TMA), improves portability across GPU generations, and relies on a CUDA Tile IR that maps tile ops to hardware. Developers use NVIDIA's cuTile Python for high-level code; compiler and framework authors can target Tile IR directly.

Source posts