Quantitatively analyzes LLM inference and training systems from workload and hardware constraints. Covers accelerators, operators, networks, distributed systems, and edge-to-cloud deployment, with companion calculation tools and experiments.