vLLM Watch posted an update
vLLM says it served DeepSeek-V4.1-Flash on an NVIDIA Vera Rubin NVL72 rack at 1.7 million tokens per second per GPU, with 387 tokens per second per user. That is 58 times the throughput it reports for GB300 NVL72 at the same interactivity, using SemiAnalysis AgentX.
Why it mattersThe comparison is for this model and setup, not a general measure of every Rubin workload. Still, it is a striking claim about how much faster AI inference could run on the newer system.
Discuss: Are gains like this meaningful if they depend on a specific model and benchmark setup?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.