I just found one of the most useful resources for understanding GPUs. No more jumping between random docs, PDFs, and forum threads… → https://modal.com/gpu-glossary/readme… Most engineers working with GPUs run into the same problem: • Concepts are scattered across different sources • Low-level details don’t connect to high-level abstractions • You end up memorizing terms without really understanding them Things like: • What is a Streaming Multiprocessor? • How does Compute Capability affect your code? • What do nvcc flags do under the hood? So the team at @modal built a GPU Glossary that connects everything in one place. And: 1. It spans the entire stack 2. It’s built for engineers 3. It’s fully interlinked The process is simple: 1. Pick a concept you don’t fully understand 2. Follow the links across the stack 3. Build intuition instead of memorizing terms If you’re working with LLMs, training, or inference at scale, this kind of understanding matters. Because GPU bottlenecks are rarely obvious from the surface. Repo here: https://modal.com/gpu-glossary/readme…

README | GPU Glossary

Source: https://modal.com/gpu-glossary/readme

██████╗ ██████╗ ██╗   ██╗
██╔════╝ ██╔══██╗██║   ██║
██║  ███╗██████╔╝██║   ██║
██║   ██║██╔═══╝ ██║   ██║
╚██████╔╝██║     ╚██████╔╝
 ╚═════╝ ╚═╝      ╚═════╝
 ██████╗ ██╗      ██████╗ ███████╗███████╗ █████╗ ██████╗ ██╗   ██╗
██╔════╝ ██║     ██╔═══██╗██╔════╝██╔════╝██╔══██╗██╔══██╗╚██╗ ██╔╝
██║  ███╗██║     ██║   ██║███████╗███████╗███████║██████╔╝ ╚████╔╝
██║   ██║██║     ██║   ██║╚════██║╚════██║██╔══██║██╔══██╗  ╚██╔╝
╚██████╔╝███████╗╚██████╔╝███████║███████║██║  ██║██║  ██║   ██║
 ╚═════╝ ╚══════╝ ╚═════╝ ╚══════╝╚══════╝╚═╝  ╚═╝╚═╝  ╚═╝   ╚═╝

We wrote this glossary to solve a problem we ran into working with GPUs here atModal: the documentation is fragmented, making it difficult to connect concepts at different levels of the stack, likeStreaming Multiprocessor Architecture,Compute Capability, andnvcc compiler flags.

So we’ve read thePDFs from NVIDIA, lurked in thegood Discords, and even boughtdead-tree textbooksto put together a glossary that spans the whole stack in one place.

This glossary, unlike a PDF or a Discord or a book, is ahypertext document-- all pages are inter-linked with one another, so you can jump down to read about theWarp Schedulerso you can better understand thethreadsthat you came across in the article on theCUDA programming model.

You can also read it linearly. To navigate between pages, use the arrow keys, the arrows at the bottom of each page, or the table of contents (in the sidebar on desktop or in the hamburger menu on mobile).

The source for the glossary is availableon GitHub.