adam-maj/tiny-gpu ? reverse-engineered prompt

I want to build a small, educational project that shows how a GPU works at the hardware level. It's not for graphics or gaming, but for helping people understand the core principles of parallel computing, like in modern AI accelerators.

The idea is to create a "tiny GPU" that can run basic math tasks, specifically matrix addition and matrix multiplication, by spreading the work across multiple simple processing cores at the same time. This should be a complete simulation of the hardware. We'll need to define the GPU's architecture, including a dispatcher that manages tasks, the compute cores themselves, and how they all access memory. It should have its own simple instruction set, too.

The whole point is for this to be a learning tool, so the design should be minimal and very well-documented to explain the concepts. A user should be able to simulate it and trace how instructions are executed.