(AI Hardware)
Why AI runs on GPUs and TPUs instead of the CPU in your laptop — in about half an hour.
A free, visual course on the chips behind AI — CPUs, GPUs and TPUs. Why neural networks are mostly matrix multiplication, the serial-vs-parallel split, how each chip is built, and which to use when.
Start with the workload →A handful of powerful cores with huge caches — built to do anything, and to do branchy, step-by-step work fast.
Thousands of small cores running the same instruction on different data — perfect for the matrix math of Chapter 1.
A chip that does almost nothing but matrix multiply — data flows through a grid of multiply-accumulate cells.