Forget the headache of old-school parallel programming! This talk spills the beans on block-based programming, a super cool and easy way to tackle parallel tasks. It chops up your data into little 'tiles' for threads to munch on, all while the framework does the heavy lifting, making GPU programming a breeze for languages like Python and C++.
You'll dive into NVIDIA's swanky new cuTile and Tile IR tech, see awesome examples (think LLAMA3!), and learn how this fresh approach makes building speedy, portable apps way more intuitive than ever before.
Watch on YouTube
Top comments (0)