Supermicro’s new Liquid Cooling Solution for AI Data Centres

It’s no secret that increasing AI workloads has created infrastructure challenges for data centre operators – leading power consumption and cooling capacity to emerge as critical bottlenecks in the deployment of advanced AI systems.

Traditional air-cooled facilities designed for general-purpose computing struggle to accommodate the thermal output of specialised AI accelerators, which can exceed 700 watts per chip – approximately seven times the heat generated by standard server processors a decade ago.

This thermal challenge has intensified as organisations worldwide deploy increasingly sophisticated AI models requiring massive computational clusters.

These pressures have accelerated interest in liquid cooling technologies, which offer significantly higher thermal transfer efficiency than air-based systems.

Now, Supermicro, a provider of high-performance computing infrastructure solutions, has introduced new direct liquid cooling technology and modular data centre design approaches to address thermal management challenges in artificial intelligence computing environments.

The company, which develops servers, storage systems and data centre infrastructure, has launched its DLC-2 direct liquid cooling solution, designed specifically for high-density AI computing clusters that generate significant heat during operation.

What is unique about the DLC-S system?

Liquid cooling technology transfers heat from computing components using circulating fluid rather than air, which becomes increasingly inefficient as computing density rises.

However, Supermicro’s DLC-2 system utilises specialised cold plates attached to processors, graphics processing units and other components to capture thermal output.

According to the company, the system can capture up to 98% of heat generated by GPUs, which typically constitute the highest heat-producing elements in AI computing systems.

This approach enables quieter operation with noise reduction of up to 50 decibels compared to air-cooled alternatives.

“With the expected demand for liquid-cooled data centres rising to 30% of all installations, we realised that current technologies were insufficient to cool these new AI-optimised systems,” says Charles Liang, President and CEO of Supermicro.

The DLC-2 system incorporates hardware designed for the Nvidia Blackwell B200 GPU architecture, supporting configurations with eight GPU units and two Intel Xeon 6 central processing units in a 4U server chassis format.

The cooling infrastructure also includes an in-rack coolant distribution unit capable of removing 250 kilowatts of heat per server rack.

This technology operates with inlet water temperatures up to 45°C, which reduces water consumption by approximately 40% compared to systems requiring colder input temperatures, as higher-temperature cooling loops reduce the need for supplemental cooling infrastructure.

Solutions to accelerate AI infrastructure deployment timelines

Alongside its cooling innovations, Supermicro has introduced a systematic approach to data centre design called Data Centre Building Block Solutions (DCBBS), which extends the company's modular design philosophy from individual servers to entire facility implementations.

The DCBBS methodology provides standardised architectural templates for creating liquid-cooled AI computing facilities, addressing power distribution, cooling infrastructure, network topology and physical layout considerations as an integrated solution.

“Supermicro's DCBBS enables clients to easily construct data centre infrastructure with the fastest time-to-market and time-to-online advantage, deploying as quickly as three months,” says Charles.

The approach aims to reduce deployment timelines for new AI computing capacity, which has become a constraint for organisations seeking to expand their AI capabilities amid increasing demand for training and inference capacity.

Supermicro indicates that the DCBBS approach can be implemented in both new construction and retrofitted into existing data centre facilities, providing flexibility for organisations with established infrastructure.

“Along with our DLC-2 technology, DCBBS also helps customers save up to 40% power, reducing 60% data centre footprint and decreasing 40% water consumption, all of which leads to 20% lower TCO” Charles adds.

Explore the latest edition of AI Magazine and be part of the conversation at our global conference series, Tech & AI LIVE. 

Discover all our upcoming events and secure your tickets today.

Also sign up to our free weekly newsletter for the latest insights and stories straight into your inbox.

AI Magazine is a BizClik brand