英伟达CES 2026宣布Rubin平台全面量产 开源Alpamayo模型家族瞄准自动驾驶

拉斯维加斯—英伟达（NVIDIA）创始人兼首席执行官黄仁勋1月5日在拉斯维加斯Fontainebleau为CES 2026作开场演讲，宣布新一代超级智能（SI）计算平台Rubin已进入全面量产。Rubin是英伟达首个采用六芯片极端协同设计的平台，由Rubin GPU、Vera CPU、NVLink 6交换、Spectrum-X以太网光子、ConnectX-9 SuperNIC与BlueField-4 DPU组成；其中Rubin GPU的NVFP4推理性能达50 petaflops。英伟达称，Rubin可将生成每个token的成本降至上一代平台的约十分之一，合作伙伴产品将于2026年下半年陆续上市。

面向自动驾驶，英伟达发布开源的Alpamayo模型家族：Alpamayo R1是业界首个开源的推理型视觉-语言-动作（VLA）模型，配套的AlpaSim是开源仿真蓝图，供车企与开发者训练、验证自动驾驶系统。首款搭载Alpamayo的量产车是全新Mercedes-Benz CLA，该车搭载英伟达的智能驾驶技术，将于今年内进入美国市场；这款车刚在欧洲Euro NCAP安全评测中获得五星。

英伟达同时扩大其开放模型布局，六大领域各成一线：Clara（医疗）、Earth-2（气候）、Nemotron（推理与多模态）、Cosmos（机器人与仿真）、GR00T（具身智能）与Alpamayo（自动驾驶）。其他发布还包括Inference Context Memory Storage Platform推理上下文存储平台、性能提升至2.6倍的DGX Spark，以及带6倍多帧生成的DLSS 4.5。黄仁勋还与Siemens宣布扩大合作，共建工业超级智能操作系统。

NVIDIA at CES 2026: Rubin Platform in Full Production, Open Alpamayo Models Target Self-Driving

拉斯维加斯—NVIDIA founder and CEO Jensen Huang opened CES 2026 with a talk at Fontainebleau in Las Vegas on January 5, announcing that the company's new Rubin computing platform is in full production. Rubin is NVIDIA's first extreme-codesigned platform built from six chips — the Rubin GPU, Vera CPU, NVLink 6 switch, Spectrum-X Ethernet Photonics, ConnectX-9 SuperNIC and BlueField-4 DPU — with the Rubin GPU delivering 50 petaflops of NVFP4 inference performance. NVIDIA says Rubin cuts the cost of generating each token to about one-tenth of the previous generation, with partner products arriving in the second half of 2026.

For self-driving, NVIDIA launched the open Alpamayo family of models: Alpamayo R1, the industry's first open reasoning vision-language-action (VLA) model, paired with AlpaSim, an open simulation blueprint for training and validating autonomous systems. The first production car built on Alpamayo is the all-new Mercedes-Benz CLA, which carries NVIDIA's AI-defined driving stack, reaches the U.S. market this year, and has just earned a five-star Euro NCAP safety rating.

NVIDIA also broadened its open-model push across six domains: Clara for healthcare, Earth-2 for climate, Nemotron for reasoning and multimodal models, Cosmos for robotics and simulation, GR00T for embodied intelligence and Alpamayo for autonomous driving. Other announcements included the Inference Context Memory Storage Platform, a DGX Spark performance boost to 2.6x, and DLSS 4.5 with 6X Multi Frame Generation. Huang also appeared with Siemens to expand their partnership on an industrial super-intelligence operating system.

来源：英伟达

发布：GSM·美国（发布链接）

图片来源：英伟达首席执行官黄仁勋在CES 2026演讲中展示芯片与服务器（图：英伟达官方博客） / NVIDIA CEO Jensen Huang presenting chips and servers during his CES 2026 talk (Photo: NVIDIA Blog)

GSM Global · 本篇英文版 English edition of this story

本频道最新更多 ›

跟进：特朗普称已与黄仁勋通话，黄仁勋说完全不知道对方是间谍，「我真的相信他」2026-11-09跟进：黄仁勋回复期限11月11日将至 英伟达尚未确认他是否出席国会闭门简报2026-11-09

← 返回科技　全部新闻　关于我们