Nvidia 揭曉強大的全新 Rubin 運算平台
Nvidia reveals the powerful new 'Rubin' computing platform
在 2026 年的國際消費電子展（Consumer Electronics Show）上，Nvidia 發布了其最新的科技傑作：Rubin 運算平台。
At the 2026 Consumer Electronics Show, Nvidia unveiled its latest technological marvel: the Rubin computing platform.
Rubin 旨在取代 Blackwell 架構，代表了一種向「代理型 AI」（agentic AI）的基礎性轉變——即具備複雜多步驟推理能力的系統。
Designed to succeed the Blackwell architecture, Rubin represents a foundational shift toward 'agentic AI'—systems capable of complex multi-step reasoning.
與傳統設計不同，Rubin 採用了「極致協同設計」（extreme codesign）方法，將包括 Rubin GPU 和 Vera CPU 在內的六大核心組件整合至一個統一的機架級系統中。
Unlike traditional designs, Rubin utilizes an 'extreme codesign' approach, integrating six core components, including the Rubin GPU and the Vera CPU, into a unified, rack-scale system.
Nvidia 報告指出，與前代產品相比，Rubin 的訓練速度提升了 3.5 倍，推理速度提升了 5 倍，同時將推理成本降低了十倍。
Nvidia reports that Rubin delivers up to 3.5 times faster training and 5 times faster inference compared to its predecessor, while also reducing inference costs by a factor of ten.
該平台以天文學家薇拉·魯賓（Vera Rubin）的名字命名，目前已進入生產階段，預計 AWS、Microsoft 和 OpenAI 等主要合作夥伴將於 2026 年下半年推出相關系統。
Named after astronomer Vera Rubin, the platform is currently in production, with major partners like AWS, Microsoft, and OpenAI expected to roll out systems in the second half of 2026.
歸根結底，Rubin 的目標是使大型智慧 AI 模型對於全球企業而言更具經濟效益與擴展性。
Ultimately, Rubin aims to make large-scale, intelligent AI models more economical and scalable for global enterprises.
