Anatomy

Machines, taken apart

The hardware under language models, taken apart one part at a time, with what each part does and why it matters.

Anatomy · 01
Anatomy of a GPU
A graphics card in four layers and ten parts: the chip, its 24 GB of memory, power delivery and the cooler, and which of them decide how fast an LLM runs.
interactive · 10 parts
More anatomies on the way.
Anatomy — fp16.in