Nvidia Stuns with First Vera Rubin Test, DeepSeek Throughput Surges 30-Fold!

Nvidia unveiled first on-chip benchmark data for Vera Rubin NVL72, paired with DeepSeek-V4-Pro on Agent coding tasks, showing up to a 30-fold surge in per-megawatt throughput versus GB300, with token costs plunging 35-fold. On the same day, the Agent-designed Vera CPU was deployed at scale by SpaceXAI, with plans to send it to space by 2028. Groq accelerators entered full mass production.