Supermicro Expands NVIDIA Blackwell Portfolio with Advanced Liquid-Cooled HGX B300 Systems
Super Micro Computer, Inc. (SMCI), a leader in IT solutions for AI, cloud, storage, and 5G/edge computing, has announced the expansion of its NVIDIA Blackwell architecture portfolio. The company is now offering new 4U and 2-OU (OCP) liquid-cooled NVIDIA HGX B300 systems, designed to deliver exceptional GPU density and energy efficiency for hyperscale data centers and AI factory deployments.
Unmatched GPU Density and Power Efficiency for AI Infrastructure
The latest additions to Supermicro’s Data Center Building Block Solutions (DCBBS) are engineered to meet the surging global demand for high-performance AI infrastructure. The new liquid-cooled NVIDIA HGX B300 systems provide up to 144 GPUs in a single rack, setting a new standard for compactness and performance density. Leveraging direct liquid-cooling technology, these systems significantly reduce power consumption and cooling costs, enabling organizations to deploy AI at scale with faster time-to-market and maximum performance per watt.
Innovative 2-OU (OCP) Liquid-Cooled System for Hyperscale Deployments
The 2-OU (OCP) liquid-cooled NVIDIA HGX B300 system is built to the 21-inch OCP Open Rack V3 (ORV3) specification, allowing hyperscale and cloud providers to maximize GPU density without sacrificing serviceability. This rack-scale solution features blind-mate manifold connections, a modular GPU/CPU tray architecture, and advanced component liquid cooling. Each system supports eight NVIDIA Blackwell Ultra GPUs at up to 1,100 W TDP, while minimizing rack footprint and energy usage. A single ORV3 rack can accommodate up to 18 nodes, totaling 144 GPUs, and scales efficiently with NVIDIA Quantum-X800 InfiniBand switches and Supermicro’s 1.8MW in-row coolant distribution units (CDUs). A SuperCluster scalable unit can be formed by combining eight compute racks, three networking racks, and two in-row CDUs, delivering a total of 1,152 GPUs.
4U Front I/O HGX B300 System for Traditional Rack Environments
For organizations utilizing traditional 19-inch EIA rack form factors, the 4U Front I/O HGX B300 Liquid-Cooled System offers the same high compute performance. This system incorporates Supermicro’s DLC-2 technology, which captures up to 98% of system-generated heat through liquid cooling. The result is superior power efficiency, reduced noise, and enhanced serviceability—ideal for dense AI training and inference clusters.
Performance and Efficiency for Next-Generation AI Workloads
Supermicro’s NVIDIA HGX B300 systems are designed to accelerate demanding AI workloads. Each system is equipped with 2.1 TB of HBM3e GPU memory, supporting larger model sizes and complex computations. Both the 2-OU (OCP) and 4U platforms deliver significant cluster-level performance improvements by doubling compute fabric network throughput to 800 Gb/s, thanks to integrated NVIDIA ConnectX-8 SuperNICs and compatibility with NVIDIA Quantum-X800 InfiniBand or NVIDIA Spectrum-4 Ethernet. These advancements are particularly beneficial for agentic AI applications, foundation model training, and large-scale multimodal inference in AI factories.
Optimized for Total Cost of Ownership and Sustainability
Supermicro developed these platforms with a focus on total cost of ownership (TCO), serviceability, and operational efficiency. The DLC-2 technology stack enables data centers to achieve up to 40% power savings and supports 45°C warm water operation, reducing water consumption and eliminating the need for chilled water and compressors. Supermicro delivers these systems as fully validated, tested racks (L11 and L12 solutions), ensuring rapid deployment for hyperscale, enterprise, and federal customers.
Comprehensive NVIDIA Blackwell Platform Portfolio
These new systems further expand Supermicro’s extensive range of NVIDIA Blackwell platforms, which includes the NVIDIA GB300 NVL72, NVIDIA HGX B200, and NVIDIA RTX PRO 6000 Blackwell Server Edition. All Supermicro NVIDIA-Certified Systems are rigorously tested to ensure optimal performance across diverse AI applications and use cases. With integrated NVIDIA networking and AI software solutions such as NVIDIA AI Enterprise and NVIDIA Run:ai, customers gain the flexibility to build scalable AI infrastructure—from single nodes to full-scale AI factories.
