cuTile Python — tiled GPU kernels in Colab for vector and matrix ops
NVIDIA cuTile Python brings tile-based GPU programming to Python, letting you write CUDA-style kernels without C++. A new tutorial walks through setup in Colab and three concrete kernels: vector addition, matrix addition, and matrix multiplication.
• Pre-flight checks for GPU, driver, CUDA, and cuTile availability
• Each kernel includes PyTorch fallback so notebooks stay runnable
• Correctness validated against PyTorch reference; median runtimes benchmarked at each stage