Hacker News Reader

As a person who finds CUDA extremely easy to write and integrate, what does Triton have to offer?

block level rather than thread level programming, automatic optimization across hyperparameters, makes it much easier to write fast kernels