Digital Event Horizon
NVIDIA's Rubin platform has revolutionized AI computing with its groundbreaking technology, accelerating any AI workload and reducing inference token cost by up to 10x. The DGX SuperPOD serves as a blueprint for large-scale deployments built on this innovative platform.
NVIDIA unveils its latest innovation – the Rubin platform, which accelerates agentic AI and mixture-of-experts (MoE) models. The Rubin platform introduces several key innovations that will drive a step-function increase in intelligence and efficiency. The platform features six new chips: NVIDIA Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch. The Rubin platform accelerates training and reduces the cost of inference token generation. It achieves this through sixth-generation NVIDIA NVLink, NVIDIA Vera CPU, NVIDIA Rubin GPU, third-generation NVIDIA Confidential Computing, and Second-generation RAS Engine. The platform is designed to accelerate any AI workload, from inference and training to long-context reasoning. NVIDIA has introduced the DGX SuperPOD, a system that will serve as the blueprint for large-scale deployments built on the Rubin platform.
NVIDIA, a pioneer in artificial intelligence (AI) computing technology, has unveiled its latest innovation – the Rubin platform. This groundbreaking platform represents an extreme codesign with NVIDIA's Rubin platform, showcasing the company's unwavering commitment to pushing the boundaries of AI computing.
The Rubin platform is engineered to accelerate agentic AI, mixture-of-experts (MoE) models, and long-context reasoning. By masterfully harnessing congestion control and performance isolation, NVIDIA has created a holistic approach that ensures as AI models grow in complexity, the networking fabric of the AI factory remains a catalyst for speed rather than a constraint.
This significant advancement is made possible by the integration of six new chips – the NVIDIA Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch. Through an advanced codesign approach, these components work in harmony to accelerate training and reduce the cost of inference token generation.
The Rubin platform introduces several key innovations that will drive a step-function increase in intelligence and efficiency:
1. Sixth-generation NVIDIA NVLink – Delivering 3.6TB/s per GPU and 260TB/s per Vera Rubin NVL72 rack, this technology enables massive MoE and long-context workloads.
2. NVIDIA Vera CPU – Featuring 88 NVIDIA custom Olympus cores, full Armv9.2 compatibility, and ultrafast NVLink-C2C connectivity for industry-leading efficient AI factory compute.
3. NVIDIA Rubin GPU – Providing 50 petaflops of NVFP4 compute for AI inference featuring a third-generation Transformer Engine with hardware-accelerated compression.
4. Third-generation NVIDIA Confidential Computing – Vera Rubin NVL72 is the first rack-scale platform delivering NVIDIA Confidential Computing, which maintains data security across CPU, GPU, and NVLink domains.
5. Second-generation RAS Engine – Spanning GPU, CPU, and NVLink, this innovation delivers real-time health monitoring, fault tolerance, and proactive maintenance, with modular cable-free trays enabling 3x faster servicing.
The Rubin platform is designed to accelerate any AI workload, from inference and training to long-context reasoning. This significant advancement has resulted in a critical milestone – a reduction of up to 10x in inference token cost compared to the previous generation.
To bring this innovation to life, NVIDIA has introduced the DGX SuperPOD, a system that will serve as the blueprint for large-scale deployments built on the Rubin platform. The DGX SuperPOD integrates:
1. NVIDIA DGX Vera Rubin NVL72 or DGX Rubin NVL8 systems
2. NVIDIA BlueField-4 DPUs for secure, software-defined infrastructure
3. NVIDIA Inference Context Memory Storage Platform for next-generation inference
4. NVIDIA ConnectX-9 SuperNICs
5. NVIDIA Quantum-X800 InfiniBand and NVIDIA Spectrum-X Ethernet
The DGX Vera Rubin NVL72 system, comprising eight systems with 576 Rubin GPUs, delivers 28.8 exaflops of FP4 performance and 600TB of fast memory. Each DGX Vera Rubin NVL72 system enables a unified memory and compute space across the rack, eliminating the need for model partitioning.
The DGX Rubin NVL8 system, featuring 64 systems with 512 Rubin GPUs, brings Rubin performance into a liquid-cooled form factor with x86 CPUs, making it an efficient on-ramp to the Rubin era for any AI project in the develop-to-deploy pipeline. Powered by eight NVIDIA Rubin GPUs and sixth-generation NVLink, each DGX Rubin NVL8 delivers 5.5x NVFP4 FLOPS compared with NVIDIA Blackwell systems.
The Rubin platform redefines the data center as a high-performance AI factory with revolutionary networking. This is achieved through:
1. Next-generation 800Gb/s end-to-end networking suite
2. Two purpose-built paths for AI infrastructure – InfiniBand and Ethernet
3. NVIDIA Quantum-X800 InfiniBand delivering industry-lowest latency and highest performance for dedicated AI clusters
4. NVIDIA Spectrum-X Ethernet providing predictable, high-performance scale-out and scale-across connectivity to AI factories using standard Ethernet protocols
Optimized connectivity for massive-scale clusters is ensured through the integration of these innovations into the DGX SuperPOD.
In conclusion, NVIDIA's Rubin platform represents a significant leap forward in AI computing. By harnessing advanced codesign and innovative technologies, NVIDIA has created a system that accelerates any AI workload, reduces inference token cost, and enables massive MoE and long-context workloads. The DGX SuperPOD serves as the blueprint for large-scale deployments built on the Rubin platform, making it an essential component in the pursuit of next-generation AI applications.
Related Information:
https://www.digitaleventhorizon.com/articles/NVIDIA-Revolutionizes-AI-Computing-with-Rubin-Platform-A-Blueprint-for-the-Next-Gen-AI-Factories-deh.shtml
https://blogs.nvidia.com/blog/dgx-superpod-rubin/
Published: Sat Jan 17 06:22:49 2026 by llama3.2 3B Q4_K_M
