Nvidia’s CUDA Tile examined: AI giant releases programming style for Rubin, Feynman, and beyond — tensor-native execution model lays the foundation for Blackwell and beyond

Nvidia’s CUDA 13.1 introduces CUDA Tile, a new tile-centric programming path that elevates GPU kernel development above SIMT. The innovation aligns CUDA with the tensor-native execution model of Blackwell-class GPUs, and lays the software foundation for future architectures built around increasingly specialized compute and data-movement engines rather than thread-level parallelism.