Two months ago, we released the GPU Glossary, an interlinked compendium of short documents about GPU programming topics.

Today, we’re announcing that the GPU Glossary is now available “open source” — on GitHub under a Creative Commons BY 4.0 License.

Why write a GPU Glossary?

Programming GPUs is hard.

In some way’s that’s fundamental, because GPUs are used for high-performance computing and high-performance computing is hard — performance breaks abstractions, performance is hard to debug, performance is relative and competitive.

But it’s also harder than it needs to be, because there isn’t the same depth and quality of reference and educational material available online. In particular, there aren’t many resources that cross the entire stack, weaving a coherent picture, the way a high-quality textbook or course would.

We wrote the GPU Glossary to help address this gap — to provide “GPU documentation for humans”.

Why make it open source?

When we shared the GPU Glossary, we observed a ton of excitement and enthusiasm from the community, from the GPU MODE Discord to Hacker News and Twitter.

We also received multiple requests to contribute — typo fixes, small corrections, entire new sections, compilation to eBook format, you name it. We accepted those we could (and thanked the contributors here), but our throughput over email and social media was limited.

That’s why we’re releasing the raw material on GitHub and under a permissive license that encourages others to contribute and makes it easier for us to handle contributions. It also allows you to fork it for your own purposes. You just need to link back to us — a bit of credit goes a long way in the attention economy!

How can I contribute?

We’ve pre-populated the GitHub Issues for the repo with some things we’d like to do to extend the glossary — like adding two sample matmul kernels, writing about Thread Block Clusters, and including a script that outputs the glossary a single Markdown file. We’d be happy to accept your contribution!