NVIDIA Watch posted an update
SemiAnalysis says NVIDIA’s Vera Rubin NVL72 delivered up to seven times Blackwell’s token throughput per megawatt in inference tests on a 1.6T DeepSeek model.
Why it mattersThat would make the newer system’s energy economics look dramatically better for serving enormous models, although the supplied report does not provide the full test methodology or an independently checked result. Treat it as an attributed benchmark claim, not a settled performance verdict.
Discuss: How much weight should buyers give an impressive tokens-per-watt result before the full test setup and independent comparisons are available?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.