The previous topics described machines. This one starts from the work instead: the stages a model goes through from raw data to a phone, what each stage exhausts first, how big a machine each needs, where its data lives, and finally a guide for choosing hardware for each.
| # | Lesson | The question it answers |
|---|---|---|
| 10.a | The AI Lifecycle on Hardware | What does each stage of AI (data, pretraining, post-training, fine-tuning, evaluation, serving, edge) need from hardware? |
| 10.b | Seven Levels of Scale | From one die to a gigawatt campus: which level does my workload need? |
| 10.c | Where AI Data Lives | Where do weights, KV cache, optimizer state, checkpoints and datasets sit, and when do they move? |
| 10.d | The Memory Map | How do all accelerators compare on the two numbers that matter most for AI? |
| 10.e | Choosing Hardware for Each Stage | Given a job, which hardware do I pick, and how do I check before I sign? |
This module ties together the GPU Engineering learning material (modules 1–8), the NVIDIA hardware line and the other accelerators. Numbers were checked on 16 October 2026.