The author discusses the complexity and mystique surrounding modern GPUs and the need for a better understanding of how they work. They highlight that while CPUs are often complex, much of that complexity is hidden from users, whereas GPUs expose a lot of complexity due to their parallel computing nature. The author also mentions the barriers to learning about GPU architectures, including architectural variations, inconsistent terminology, and software abstractions. They emphasize the importance of cutting through the mystery and provide a basic explanation of recent architectures from Nvidia and AMD. The author also includes references and links for further reading on the topic.
https://thechipletter.substack.com/p/demystifying-gpu-compute-architectures