Supermicro unveils new liquid-cooled and air-cooled NVIDIA Blackwell data center systems

Supermicro has announced the expansion of its NVIDIA Blackwell system portfolio with two new front input/output (I/O) systems designed for high-density artificial intelligence (AI) and cloud-scale data center deployments. The new models include a 4U DLC-2 liquid-cooled system, which Supermicro claims provides up to 40 percent data center power savings, and an 8U air-cooled system optimized for cold aisle serviceability and flexible memory configurations.

The 4U DLC-2 liquid-cooled platform supports the NVIDIA HGX B200 architecture and is ready for volume shipment. According to Supermicro, the next-generation Direct Liquid Cooling design enables significant reductions in both power and water consumption—up to 40 percent less compared to traditional systems—and allows for warm water cooling at inlet temperatures up to 45°C. System-level heat capture by liquid cooling can reach up to 98 percent, as it cools all major components, including processors, graphics processing units (GPUs), memory modules, power supplies, and peripheral switches. The liquid-cooled system is also designed for quiet data center operations, with noise levels as low as 50 dB.

Both the 4U liquid-cooled and 8U air-cooled front I/O systems feature support for eight NVIDIA ConnectX-7 400G network interface cards (NICs) and two NVIDIA Bluefield-3 data processing units (DPUs), all relocated to the front of the chassis for easier cabling and cold aisle access. These systems support NVIDIA Quantum-2 InfiniBand and Spectrum-X Ethernet for high-bandwidth compute fabrics. The new platforms also offer 32 dual in-line memory module (DIMM) slots, enabling up to 8 TB capacity at 5200 MT/s or up to 4 TB at 6400 MT/s with DDR5 RDIMM, and can operate with dual-socket Intel Xeon 6700 Series processors with a thermal design power up to 350 W.

The 8U air-cooled model is engineered for environments without liquid cooling infrastructure. It maintains a compact footprint by incorporating a reduced-height CPU tray, while retaining a full-height 6U GPU tray to maximize airflow and thermal performance. Both systems are compatible with NVIDIA HGX B200 eight-GPU configurations, offering 1.4 TB total HBM3e GPU memory per appliance and GPU interconnects via fifth-generation NVLink at 1.8 TB per second.

According to Supermicro, the front I/O systems allow for easier deployment, improved cabling, and lower operational expenses in large-scale AI training clusters or inference environments, and will support future NVIDIA HGX B300 platforms. These systems are intended for hyperscale data centers, colocation providers, and technology vendors building dense, scalable AI factories.

“Supermicro’s DLC-2 enabled NVIDIA HGX B200 system leads our portfolio to achieve greater power savings and faster time to online for AI Factory deployments,” said Charles Liang, CEO and president, Supermicro. “Our Building Block architecture enables us to quickly deliver solutions exactly as our customers request. Supermicro’s extensive portfolio now can offer precisely optimized NVIDIA Blackwell solutions to a diverse range of AI infrastructure environments, whether deploying into an air- or liquid-cooled facility.”

“Based on the latest NVIDIA Blackwell architecture, Supermicro’s new front I/O B200 systems equip enterprises to deploy and scale AI at unprecedented speed—delivering breakthrough innovation, efficiency, and operational excellence,” said Kaustubh Sanghani, vice president of GPU product management at NVIDIA.

Source: Supermicro

Get Data Center Engineering News In Your Inbox:

Popular Posts:

GmyybkpD3o7f45xka2pctu42nog
FS launches high-density 1U WDM platform for data centers, maximizing rack space and passive cooling
Screenshot
Knowles launches safety-certified high-voltage ceramic capacitors for data center applications
kbra-logo-new
KBRA assigns preliminary ratings to colocation data center ABS backed by 32 facilitie
pr-image-forcadence-and-samsung-foundry-expand-collaboration-to-accelerate-ai-data-center-chip-design-and-power-optimization-jpg
Cadence and Samsung Foundry expand collaboration to accelerate AI data center chip design and power optimization
Yokogawa-T&M---SL2000-high-speed-data-acquisition-unit
Yokogawa releases SL2000 high-speed data acquisition unit for multi-channel measurement and analysis applications

Share Your Data Center Engineering News

Do you have a new product announcement, webinar, whitepaper, or article topic? 

Get Data Center Engineering News In Your Inbox: