Code for the Kernel Engineering Workshop by Vizuara. Each lecture has its own folder with the programs we run in class, so you can repeat the measurements from the lecture on a GPU of your own.
The GPU work runs on Modal, a cloud service that rents you a GPU by the second. Your laptop only sends the code and shows the results, so you need a Modal account and Python 3 on your laptop (we tested with Python 3.11), not a GPU. New Modal accounts get free monthly credits; see modal.com/pricing.
| Folder | Lecture | What you run |
|---|---|---|
lecture1 |
Lecture 1: How fast can this go? | Measure an H100's two speed limits, serve Qwen3-8B on it with vLLM, and chat with your model from a browser. |
lecture3 |
Lecture 3: How does a GPU run your code? | Time the lecture's CUDA kernels on a photo of a dog on one H100, and count the warps whose 32 threads disagree at an if. |
assignment1 |
Assignment 1: Rooflines, arithmetic intensity and the memory hierarchy | The handout and its four experiments on one H100: the GPU's two ceilings and its L2, the roofline traced with real matmuls, coalescing and bytes in flight, and RoPE four ways. |
We add a folder here after each lecture is taught.