NVIDIA s latest AI server chip

Nvidia's latest AI server platform, Rubin, integrates six next-generation chips—including Vera CPUs and Rubin GPUs—designed for large-scale AI training and inference with unprecedented efficiency and performance.Rubin Platform OverviewNvidia's Rubin platform is a next-generation AI supercomputing solution that combines six specialized chips into a tightly integrated system for AI workloads . The platform is designed for agentic AI, advanced reasoning, and mixture-of-experts (MoE) models, enabling multi-step inference over extremely long contexts with high efficiency . Key components include:NVIDIA Vera CPU: 88 custom Olympus cores optimized for AI data orchestration, Arm-compatible, and designed to manage data movement, memory, and control flow at AI factory scale .NVIDIA Rubin GPU: High-performance AI compute with HBM4 memory and the new Transformer Engine for accelerated model training and inference .NVIDIA NVLink 6 Switch: Sixth-generation interconnect providing 3.6 TB/s GPU-to-GPU bandwidth for high-speed communication .NVIDIA ConnectX-9 SuperNIC: Low-latency, high-throughput networking interface for scale-out AI deployments .NVIDIA BlueField-4 DPU: Infrastructure offload and security processor to manage AI workloads efficiently .NVIDIA Spectrum-6 Ethernet Switch: Co-packaged optics for reliable, scale-out connectivity .Performance and EfficiencyThe Rubin platform achieves up to 10x lower cost per token compared to the previous Blackwell platform and can train MoE models with 4x fewer GPUs, significantly reducing infrastructure requirements . The NVL72 rack-scale system combines 72 GPUs into a single server, and multiple NVL72s can be linked to form a DGX SuperPOD, creating massive AI supercomputers suitable for hyperscalers .Industry AdoptionMajor companies are rapidly adopting Rubin systems. Meta plans to deploy millions of Nvidia chips, including standalone Vera CPUs and next-generation GPUs, across its AI data centers to support AI features on platforms like WhatsApp . Microsoft and CoreWeave are also among the first to implement Rubin systems for large-scale AI workloads . These deployments highlight Rubin's role in enabling mainstream AI adoption at hyperscale.Additional InnovationsThe Rubin platform introduces several breakthroughs:Transformer Engine for optimized AI model training.Confidential Computing and RAS Engine for secure, reliable operations.Advanced NVLink interconnects for high-speed GPU communication.Integration of CPUs, GPUs, networking, and storage into a co-designed AI factory architecture .SummaryNvidia's Rubin platform represents a major leap in AI server technology, combining next-generation CPUs, GPUs, and networking chips into a unified system that maximizes performance, efficiency, and scalability. It is designed to accelerate AI training and inference, reduce costs, and support the growing demand for large-scale AI applications across hyperscalers and enterprise data centers .