Air & Liquid-Cooled B300 Systems
Supermicro NVIDIA HGX™ B300 8-GPU Systems, 8U Air-Cooled and 4U Liquid-Cooled
As AI models grow larger and more complex, infrastructure must deliver extreme compute density, memory bandwidth, and networking performance—without sacrificing efficiency or scalability. This datasheet introduces Supermicro NVIDIA HGX™ B300 8-GPU systems, powered by NVIDIA Blackwell Ultra architecture, designed as the core building blocks of the world’s largest AI clusters.
Whether organisations prioritise simplified air cooling or maximum density through liquid cooling, the datasheet shows how these systems scale seamlessly into fully integrated 72-node clusters, supporting hundreds of GPUs per rack with non-blocking, high-speed fabrics and enterprise-grade power and cooling design.
Key Insights Include:
- Blackwell Ultra delivers a new performance tier for AI — NVIDIA HGX™ B300 platforms provide massive GPU memory, ultra-fast NVLink interconnects, and system-level optimisation for the largest AI models and most demanding workloads.
- Choice of air or liquid cooling enables deployment flexibility — Organisations can choose 8U air-cooled systems for simpler integration or 4U liquid-cooled systems for higher density and thermal efficiency at scale.
- Integrated high-speed networking keeps GPUs fully utilised — Built-in NVIDIA ConnectX®-8 SuperNICs and BlueField®-3 DPUs support up to 800Gb/s, enabling non-blocking fabrics and GPUDirect RDMA for large-scale AI clusters.
- Rack-scale designs simplify ultra-scale AI deployment — Pre-defined air-cooled and liquid-cooled 72-node cluster configurations support up to 576 NVIDIA B300 GPUs, with integrated management, power, and fabric design.
- Engineered for efficiency, reliability, and growth — Redundant Titanium-level power supplies, hot-swap NVMe storage, advanced cooling infrastructure, and modular rack designs ensure predictable performance and long-term scalability.
Download the datasheet to explore system specifications, cluster architectures, and deployment options for NVIDIA Blackwell-powered Supermicro HGX™ B300 solutions.
White Paper from 