H100 vs GB200 NVL72 Training Benchmarks – Power, TCO, and Reliability Analysis, Software Improvement Over Time
Source
Dylan Patel
Author
Dylan Patel
Date
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Why it matters
Nvidia's generational claims collapse once downtime and lost engineering time enter the math; these are measured MFU and cost-per-tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → figures across cluster scales and software versions you can reason about directly.
articleNVIDIA Exemplar Cloud: Lessons for Unlocking Full Performance on AI InfrastructureElizabeth Goodman
articleVera Rubin NVL72 vs GB200 NVL72? Inference TCO & Architecture AnalysisAlec Ibarra
articleNVIDIA Vera Rubin and Blackwell Set a New Standard for Agentic AI Performance per WattElizabeth Goodman