NVIDIA DGX SuperPOD 为基于 Rubin 的系统奠定基础

NVIDIA DGX SuperPOD为基于Rubin的系统奠定基础 NVIDIA DGX Rubin systems unify the latest NVIDIA breakthroughs in compute, networking and software to deliver up to10x reduction in inference token cost compared with the NVIDIA Blackwell platform — accelerating any AI workload,from inference and training to long-context reasoning. NVIDIA DGX Rubin系统整合了NVIDIA在计算、⽹络和软件领域的最新突破，与NVIDIA Blackwell平台相⽐，可将推理token成本降低多达10倍——加速从推理和训练到⻓上下⽂推理的各类AI⼯作负载。 January 5, 2026byCharlie Boyle2026年1⽉5⽇，作者：Charlie Boyle NVIDIA DGX SuperPOD is paving the way for large-scale system deployments built on the NVIDIA Rubinplatform — the next leap forward in AI computing. NVIDIA DGX SuperPOD正在为基于NVIDIA Rubin平台的⼤规模系统部署铺平道路——这是AI计算领域的下⼀次⻜跃。 At the CES trade show in Las Vegas,NVIDIA today introduced the Rubin platform,comprising six new chipsdesigned to deliver one incredible AI supercomputer, and engineered to accelerate agentic AI,mixture-of-experts (MoE) models and long-context reasoning. 在拉斯维加斯举⾏的CES展会上，NVIDIA今⽇发布了Rubin平台，该平台由六款全新芯⽚组成，旨在打造⼀台卓越的AI超级计算机，并经过专⻔设计以加速代理式AI、专家混合（MoE）模型以及⻓上下⽂推理。 The Rubin platform unites six chips — theNVIDIA Vera CPU, Rubin GPU,NVLink 6 Switch, ConnectX-9SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet Switch — through an advanced codesign approachthat accelerates training and reduces the cost of inference token generation. Rubin平台通过⼀种先进的协同设计⽅法，将六款芯⽚——NVIDIA Vera CPU、Rubin GPU、NVLink 6Switch、ConnectX-9 SuperNIC、BlueField-4 DPU和Spectrum-6以太⽹交换机——整合在⼀起，加速训练并降低推理阶段⽣成token的成本。 DGX SuperPODremains the foundational design for deploying Rubin-based systems across enterprise andresearch environments. DGX SuperPOD仍然是跨企业和研究环境部署基于Rubin系统的基础性设计。 TheNVIDIA DGX platformaddresses the entire technology stack — from NVIDIA computing to networkingto software — as a single, cohesive system, removing the burden of infrastructure integration and allowingteams to focus on AI innovation and business results. NVIDIA DGX平台将从NVIDIA计算、⽹络到软件的整个技术栈整合为⼀个统⼀、协同的系统，消除了基础设施集成的负担，使团队能够专注于AI创新和业务成果。 “Rubin arrives at exactly the right moment, as AI computing demand for both training and inference is goingthrough the roof,” said Jensen Huang, founder and CEO of NVIDIA. “Rubin在恰到好处的时刻到来，因为⽆论是训练还是推理，AI计算需求都正呈爆炸式增⻓，”NVIDIA创始⼈兼⾸席执⾏官⻩仁勋表示。 New Platform for the AI Industrial Revolution ⼈⼯智能⼯业⾰命的新平台 The Rubin platformused in the new DGX systems introduces five major technology advancements designedto drive a step-function increase in intelligence and efficiency: 新⼀代DGX系统所采⽤的Rubin平台引⼊了五项重⼤技术进步，旨在推动智能⽔平和效率实现阶跃式提升： Sixth-Generation NVIDIA NVLink— 3.6TB/s per GPU and 260TB/s per Vera Rubin NVL72 rack formassive MoE and long-context workloads. 第六代NVIDIA NVLink ——每块GPU 3.6TB/s、每个Vera Rubin NVL72机架260TB/s，专为⼤规模MoE和⻓上下⽂⼯作负载⽽打造。 NVIDIA Vera CPU— 88 NVIDIA custom Olympus cores, full Armv9.2 compatibility and ultrafast NVLink-C2C connectivity for industry-leading efficient AI factory compute. NVIDIA Vera CPU ——拥有88个NVIDIA⾃定义Olympus核⼼，完全兼容Armv9.2，并具备超⾼速NVLink-C2C互连，为⾏业领先的⾼效AI⼯⼚计算提供⽀持。 NVIDIA Rubin GPU— 50 petaflops of NVFP4 compute for AI inference featuring a third-generationTransformer Engine with hardware-accelerated compression. NVIDIA Rubin GPU ——提供50拍浮点的NVFP4 AI推理算⼒，配备第三代Transformer Engine，并⽀持硬件加速压缩。 Third-Generation NVIDIA Confidential Computing— Vera Rubin NVL72 is the first rack-scale platformdelivering NVIDIA Confidential Computing, which maintains data security across CPU, GPU and NVLinkdomains. 第三代NVIDIA机密计算—— Vera Rubin NVL72是⾸个在机架级别提供NVIDIA机密计算的平台，可在CPU、GPU和NVLink域之间保持数据安全。 Second-Generation RAS Engine— Spanning GPU, CPU and NVLink, the NVIDIA Rubin platformdelivers real-time health monitoring, fault tolerance and proactive maintenance, with modular cable-freetrays enabling 3x faster servicing. 第⼆代RAS引擎——覆盖GPU、CPU和NVLink，NVIDIA Rubin平台提供实时健康监测、容错能⼒和主动维护，并通过模块化⽆缆托盘实现快3倍的维护效率。 Together, these innovations deliver up to 10x reduction in inference token cost of the previous generation —a critical milestone as AI models grow in size, context and reasoning depth. 这些创新共同实现了相⽐上⼀代⾼达10倍的推理token成本降低——这是随着AI模型在规模、上下⽂⻓度和推理深度不断增⻓⽽取得的关键⾥程碑。 DGX SuperPOD: The Blueprint for NVIDIA Rubin Scale-Out DGX SuperPOD：NVIDIA Rubin横向扩展的蓝图 Rubin-based DGX SuperPOD deployments will integrate: 基于Rubin的DGX SuperPOD部署将集成： NVIDIA DGX Vera Rubin NVL72orDGX Rubin NVL8systemsNVIDIA DGX Vera Rubin NVL72或DGX Rubin NVL8系统NVIDIA BlueField-4 DPUsfor secure, software-defined infrastructure⽤于安全、软件定义基础设施的NVIDIA BlueField-4 DPUNVIDIA Inference Context Memory Storage Platformfor next-generation inference⾯向下⼀代推理的NVIDIA推理上下⽂内存存储平台NVIDIA ConnectX-9 SuperNICsNVIDIA Quantum-X800 InfiniBandandNVIDIA Spectrum-X EthernetNVIDIA Quantum-X800 InfiniBand与NVIDIA Spectrum-X以太⽹NVIDIA Mission Controlfor automated AI infrastructure orchestration and operations⽤于⾃动化AI基础设施编排与运维的NVIDIA Mission Control NVIDIA DGX SuperPOD with DGX Vera Rubin NVL72 unifies eightDGX Vera Rubin NVL72 systems,featuring 576 Rubin GPUs, to deliver 28.8 exaflops of FP4 performance and 600TB of fast memory. EachDGX Vera