Following Supermicro's DCBBS Blueprints for NVIDIA Vera Rubin NVL72 and NVIDIA HGX Rubin NVL8 introduced at Computex, the Blueprint for HPC and AI applies the same end-to-end methodology to scientific computing. The Blueprint is based on Supermicro's DCBBS, which provide the necessary compute, networking, advanced liquid cooling, power distribution, and site infrastructure, delivered by a team of Supermicro DCBBS experts to accelerate time-to-online for research institutions and supercomputing centers.

Researchers increasingly rely on a converged approach to computing, pairing traditional FP64 double-precision simulation with accelerated computing and AI methods, revolutionizing time-to-discovery across climate research, drug discovery, materials science, and energy. The NVIDIA Vera Rubin NVL4 platform is built for this convergence, and the DCBBS Blueprint for HPC defines the steps to deploy it successfully, backed by Supermicro's proven track-record building the world's largest liquid-cooled supercomputing clusters featuring over 100,000 GPUs.

The Blueprint covers the full end-to-end sequence that Supermicro has used to complete large-scale liquid-cooled projects at record-breaking speeds. On-site facility surveys conducted by the Supermicro experts assess loading dock access, data hall measurements and clearances, floor load ratings, and existing power and cooling infrastructure to inform a design proposal tailored to each project. Solution integration begins well before delivery, with racking, stacking, cabling, and system-level (L10) and cluster-level (L11) testing performed in Supermicro's global manufacturing facilities. White-glove delivery and on-site integration cover rack placement, power and cooling connections, network cabling, commissioning, and on-site solution validation, with ongoing support options including on-site response times as fast as 4 hours for mission-critical uptime.

A Scalable AI and HPC Solution to Modernize Computing Infrastructure for Scientific Research

The Supermicro DCBBS Blueprint for HPC and AI based on the NVIDIA Vera Rubin NVL4 Scalable Unit contains the following, which can be multiplied to deploy clusters of any size, from 3.2MW to 1GW:

  • 8x liquid-cooled compute racks in customized 52U, 750mm-wide enclosures each housing 36 NVIDIA Vera Rubin NVL4 nodes within a 362 kW envelope, for a total of 288 nodes, up to 1,152 NVIDIA Rubin GPUs, and 576 NVIDIA Vera CPUs per Scalable Unit
  • Advanced Direct Liquid Cooling technology stack (DLC-2), including 3x in-row cooling distribution units (up to 1.8MW each) per Scalable Unit in a 2+1 redundant configuration, direct-to-chip copper cold plates, and vertically mounted cooling distribution manifolds, featuring Supermicro SMC PG25-A coolant engineered for exceptional chemical and thermal stability
  • NVIDIA Quantum-X800 InfiniBand compute fabric across dedicated networking switch racks with fully liquid-cooled options available, providing the high-bandwidth scale-out interconnect for distributed scientific and AI workloads
  • Rack power and management: 8x 72 kW power shelves per compute rack delivering busbar power for the 362 kW rack envelope, with two ToR management switches per rack for out-of-band control

Configurations for HPC and AI based on the NVIDIA GB200 NVL4 are also available for immediate deployment.

Supermicro DCBBS delivers complete, modular AI infrastructure built from validated components and subsystems, enabling flexible deployment from individual servers and networking to full rack-scale and data center-level solutions, including software and services. Supermicro continues to lead the industry with its comprehensive portfolio of AI infrastructure solutions, enabling organizations worldwide to deploy scalable, efficient, and environmentally responsible AI data centers.