NVIDIA CUDA Tile introduces 'tile-based parallel programming' and it's being described as a major update to the CUDA platform, which powers a lot of AI.

开发者不再需要逐线程地编写执行路径，而是把数据组织成Tile（瓦片），然后定义在这些Tile上执行什么运算。至于怎么把这些运算映射到GPU的线程、Warp和Tensor Core上，编译器和运行时会自动搞定。

很少有人像他一样，在 通用CPU、移动SoC、车载SoC、AI加速器 上都做过一线设计和架构决策。近几年他频繁在TSMC、三星等论坛谈未来工艺与架构，被称为「半导体设计传奇」。

The latest court filings in the Google versus DOJ antitrust case restricts the tech giant from making default search ...

Think of the Blind 75 as a special list of 75 coding puzzles from a website called LeetCode. It’s like a training plan for ...

While most remote jobs are getting affected, some aren't dying- they are actually evolving. Certain roles remain ...