NVIDIA Blackwell Solutions

From Supermicro

A liquid cooled AI system solution that fits inside a traditional rack

Contact Sales

News NVIDIA Blackwell Solutions Direct Liquid Cooling DLC-2 Data Center Building Block Solutions Supermicro AS-2115HV-TNRT Data Center Management Guide IoT ARS-E103-JONX

Innovative AI Infrastructure and Cooling

Supermicro's offering NVIDIA Blackwell architecture solutions, designed and built in close collaboration with NVIDIA. It's used in the world's biggest supercomputer, xAI's Colossus, to power 'Grok' and is one of the most powerful AI training system yet - with 200K GPUs, 194 Petabytes/s memory bandwidth, 3.6 TeraBits/s network bandwidth per server and more than 1 Exabyte of storage capacity. All built in just astonishingly 122 days. The solutions comes packed with customization options available for the NVIDA Blackwell architecture, but to summarize here's the most important ones.

Quick facts

Unprecedented Computational Performance
Easy maintenance and serviceability
Density
Liquid cooled
Air cooled
AI data center building blocks
Sustainable
Cutting-edge solutions
Accelerate AI workloads

Liquid cooled

The new liquid cooled 4U NVIDIA HGX B200 8-GPU system comes with many cool features, such as newly developed cold plates and advanced tubing design paired with a new 250kW coolant distribution unit (CDU) which more than doubles the cooling capacity compared to the previous generation in the same 4U form factor. The rack is easily serviceable and are available in 42U, 48U and 52U configuration. With the clever design of vertical coolant distribution manifolds (CDM) unlocks enough space for 8 systems, 64 NVIDIA Blackwell GPUs in a 42U rack and all the way up to 12 systems with 96 NVIDIA GPUs in a 52U rack.

Features

Up to 96 GPUs in a rack
1.8TB/s GPU-GPU interconnect, 180GB HBM3e per GPU
Dual Intel Xeon® 6 or AMD EPYC 9005/9004 series processors
Next-gen liquid cooling with max 250kW capacity per rack

Air cooled

The air cooled version comes as a redesigned system and is optimized for NVIDIA HGX B200 8-GPU. The new air-cooled 10U NVIDIA HGX B200 system features a redesigned chassis with expanded thermal headroom to accommodate eight 1000W TDP Blackwell GPUs. You’ll be able to fit up to 4 new 10U air-cooled systems in a rack which provides up to 15x inference and 3x training performance. All Supermicro NVIDIA HGX B200 systems are equipped with a 1:1 GPU-to-NIC ratio supporting NVIDIA BlueField®-3 or NVIDIA ConnectX®-7 for scaling across a high-performance compute fabric.

Features

Up to 32 GPUs in a rack
1.8TB/s GPU-GPU interconnect, 180GB HBM3e per GPU
Dual Intel Xeon® 6 or AMD EPYC 9005/9004 series processors
Next-gen air-cooling heatsink and enhanced serviceability

Vast Experience

Supermicros AI data center building blocks power the largest liquid cooled AI data center deployment in the world.

Flexible

Tune and dial in to fit your needs. Choose between air or liquid-cooling. The system offers GPU-optimized, multiple system and rack form factors, CPUs, storage and networking options.

Liquid Cooling Pioneer

Tested and proven, plug-and-play liquid cooling solutions that sustains the rapidly growing AI revolution. Designed specifically for NVIDIA Blackwell architecture.

Fast Time to Online

Get up and running in no time. Experience world class accelerated delivery with global capacity, one-site services, to bring your AI to production, faster than ever.