Supermicro outlines HPC blueprint for Rubin NVL4

Key Takeaways

• Supermicro introduces a Data Center Building Block Solutions (DCBBS) blueprint for High-Performance Computing (HPC) based on the NVIDIA Vera Rubin NVL4 platform
• The blueprint provides a comprehensive deployment approach for scientific computing, including compute, networking, liquid cooling, power distribution, and site infrastructure
• The NVIDIA Vera Rubin NVL4 platform is designed for mixed workloads combining FP64 double-precision simulation with accelerated computing and AI methods
• The DCBBS blueprint covers the deployment sequence for large liquid-cooled projects, including on-site facility surveys and integration before delivery

Introduction to HPC and the NVIDIA Vera Rubin NVL4 Platform

The increasing demand for scientific computing has led to the development of specialized platforms, such as the NVIDIA Vera Rubin NVL4. This platform is designed to handle mixed workloads that combine FP64 double-precision simulation with accelerated computing and AI methods. Supermicro has introduced a DCBBS blueprint for HPC based on the NVIDIA Vera Rubin NVL4 platform, providing a comprehensive deployment approach for research institutions and supercomputing centers.

HPC and AI Configuration Details

The Supermicro DCBBS blueprint for HPC and AI includes the following building blocks:

• Eight liquid-cooled compute racks with 36 NVIDIA Vera Rubin NVL4 nodes each, totaling 288 nodes, up to 1,152 NVIDIA Rubin GPUs, and 576 NVIDIA Vera CPUs per scalable unit
• A direct liquid cooling technology stack with three in-row cooling distribution units, direct-to-chip copper cold plates, and vertically mounted cooling distribution manifolds
• NVIDIA Quantum-X800 InfiniBand compute fabric across dedicated networking switch racks
• Rack power and management with eight 72-kW power shelves per compute rack and two top-of-rack management switches per rack

Comparison of HPC Platforms

Deployment and Support

The DCBBS blueprint covers the deployment sequence for large liquid-cooled projects, including:

• On-site facility surveys to evaluate loading dock access, data hall dimensions, clearances, floor load ratings, and existing power and cooling infrastructure
• Integration before delivery, including racking, stacking, cabling, and system-level and cluster-level testing at Supermicro manufacturing facilities
• On-site work, including rack placement, power and cooling connections, network cabling, commissioning, and validation, with support options that include on-site response for critical systems

Bottom Line

The Supermicro DCBBS blueprint for HPC based on the NVIDIA Vera Rubin NVL4 platform provides a comprehensive deployment approach for scientific computing, including compute, networking, liquid cooling, power distribution, and site infrastructure. With its scalable design and high-performance capabilities, this platform is well-suited for research institutions and supercomputing centers handling mixed workloads combining FP64 double-precision simulation with accelerated computing and AI methods. By following the DCBBS blueprint, organizations can ensure a successful deployment and maximize the benefits of the NVIDIA Vera Rubin NVL4 platform.