NVIDIA Vera Rubin NVL72
Supermicro's NVIDIA Vera Rubin NVL72 is designed to deliver the token throughput today's agentic AI workloads demand, targeting up to 10x the per-watt throughput of NVIDIA Blackwell. As frontier models grow, the NVIDIA Vera Rubin platform multiplies GPU memory, NVLink™, and networking bandwidth at once across 72 Rubin GPUs and 36 Vera CPUs. Supermicro's Data Center Building Block Solutions® accelerates time-to-online.
Vera Rubin NVL72
Liquid Cooled Rack-Scale Solution
Rack-Scale AI with 72 Liquid-Cooled NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs
NVIDIA Vera CPU Rack
Supermicro's air-cooled and 100% direct liquid-cooled NVIDIA Vera CPU racks are integrated systems for agentic AI, reinforcement learning, and HPC simulation. Air-cooled configurations support up to 64 Vera CPUs per 48U rack; the liquid-cooled configuration delivers 256 CPUs, up to 22,528 cores, and 300TB/s of aggregate memory bandwidth. Delivered through Supermicro DCBBS, the liquid-cooled rack combines DLC-2 cooling, third-generation NVIDIA MGX™ racks with AC or 800 VDC power options, in-row CDUs, and on-site services for faster time-to-online, with optional rear door heat exchangers and L2A sidecar CDUs for flexible facility deployment.
Vera CPU Rack
Liquid Cooled Rack-Scale Solution
High-density Liquid-cooled Rack with 256 NVIDIA Vera CPUs for Agentic AI at Scale
2U NVIDIA HGX Rubin NVL8 System
The 2U HGX Rubin NVL8 system provides the densest and most flexible HGX platform — and the first HGX platform to offer greater flexibility in CPU selections including NVIDIA Vera CPUs alongside next-generation AMD and Intel x86 processors. Built on the NVIDIA MGX rack architecture with Supermicro’s blind mate busbar and manifold for tool-free rack integration, it gives customers the freedom to pair eight Rubin GPUs with the CPU platform that best fits their workload and software stack.
2U Liquid-cooled System
For NVIDIA HGX Rubin NVL8
Front I/O Liquid-cooled system designed for the NVIDIA MGX rack architecture with Supermicro’s blindmate busbar
DCBBS Blueprints for NVIDIA Vera Rubin NVL72
Built to Scale from 5MW to 1GW
- Complete 1,152-GPU NVIDIA Vera Rubin NVL72 Scalable Unit per 5MW power envelope, multiplied to scale from a single unit to gigawatt-class AI factories
- 331 TB of HBM4 GPU memory* and 864 TB of LPDDR5X CPU memory per Scalable Unit, coherently accessible across the NVLink™ fabric
- Industry-leading DLC-2 direct liquid cooling, from direct-to-chip cold plates through 1MW cooling towers, sized for 227 kW per rack
- Dedicated Supermicro team across the full lifecycle: site survey, project design, integration, deployment, and ongoing support
- Supporting NVIDIA’s latest reference architecture integrating NVIDIA Context Memory Storage Platform, NVIDIA Spectrum™-X Ethernet, and NVIDIA Quantum-X800 InfiniBand Platform
- Management Software Suite: End-to-end SuperCloud software delivers unified infrastructure control, deployment automation, developer tools, and multi-tenant GPU cloud management
* Physical GPU memory
DCBBS Blueprints for NVIDIA Vera Rubin NVL4
Native FP64 Performance for Converged HPC and AI Infrastructure
- Complete 1,152-GPU NVIDIA Vera Rubin NVL4 Scalable, multiplied to scale from a single scalable unit to gigawatt HPC and AI deployments
- 331 TB of HBM4 GPU memory* and 864 TB of LPDDR5X CPU memory per Scalable Unit, with over 4 TB of coherent memory per NVL4 node
- Industry-leading DLC-2 direct liquid cooling sized for 360 kW per rack, cooled by 1.8 MW in-row CDUs in 2+1 redundancy
- Full-stack single-vendor solution spanning compute, storage, networking, power, cooling, and site infrastructure
- Dedicated Supermicro team across the full lifecycle: site survey, project design, integration, deployment, and ongoing support
- Optimized HPC and AI Compute fabric aligned with NVIDIA's reference architecture, featuring NVIDIA Quantum-X800 InfiniBand
- Management Software Suite: End-to-end SuperCloud software delivers unified infrastructure control, deployment automation, developer tools, and multi-tenant GPU cloud management
* Physical GPU memory
Contact Supermicro today to design your next-generation AI data center.