Down to the silicon: how chips are laid out and do their work.
One GPU core, unit by unit: schedulers, register files, ALU lanes, ray tracing and neural accelerators, caches. Run a workload.
