It seems that within the past few years, there have been a great increase in the number of articles on low-level GPU internals/programming, which was previously a very niche subject. However, once you understand that a GPU is conceptually a processor specialised for graphics operations, with its own instruction set and RAM, then everything seems quite straightforward, and you can actually start thinking about how to implement one.

The specification is a behavioral target, rather than a circuit diagram.

Did anyone else get a slight feeling of LLM-ism from that sentence? Not a complaint, because I've noticed AI is definitely leading to some great boosts (the somewhat related https://news.ycombinator.com/item?id=49717638 is on the front page as I write this.)