Core Insights - NVIDIA has launched CUDA 13.1 and CUDA Tile, marking the most significant advancement since the platform's inception approximately 20 years ago [1] Group 1: Product Features - CUDA is a parallel computing platform and programming model developed by NVIDIA, enabling developers to leverage the computational power of GPUs to enhance application performance [1] - The new tile-based programming option allows developers to write algorithms with fine control over execution, particularly beneficial for multiple GPU architectures [1] - CUDA Tile is available in Python, with plans for a C++ compatible version to be released in the future [1] Group 2: Developer Benefits - The introduction of a virtual instruction set for modular parallel programming enables higher-level algorithm writing while abstracting hardware details like tensor cores [1] - Developers can specify data blocks (tiles) for algorithm writing, allowing the compiler and runtime to manage execution without needing to set it at the element level [1] - NVIDIA aims to release the cutting-edge language of CUDA Tile as an open-source project, enhancing its integration with AI development frameworks [1]
英伟达(NVDA.US)推出CUDA 13.1 与 CUDA Tile 黄仁勋称二十年来最大升级