Cooling Down AI & Data Centers: Direct-to-Chip Cooling - WGI (2024)

• July 31, 2024
• Under Preview

Direct-To-Chip Cooling (D2C)

With the introduction of groundbreaking technologies such as the NVIDIA H100 Tensor Core GPUs, 2024 is set to become a landmark year for AI and computing advancements. However, the immense power generated by these GPUs presents significant cooling challenges… This is where Direct-To-Chip (D2C) Cooling comes into play. This innovative, waterless technology balances performance and sustainability, paving the way for the future of AI.

Direct-to-chip cold plate cooling involves circulating a cool liquid through a plate that is in direct contact with major heat sources like the CPU. Due to the advent of increasingly powerful GPUs, this method has evolved from a luxury to a necessity for high-density clusters in advanced data centers, edge computing, and federal applications.

Direct-to-chip (D2C) liquid cooling is evolving from a perceived luxury to an indispensable requirement for fueling high-density clusters in higher-end data centers, edge computing, and federal applications.

D2C liquid cooling targets the hottest components within servers. Despite its high cost and the potential risks associated with cooling system failures, it effectively manages heat by circulating liquid to the most heat-intensive components within IT equipment.

Direct-to-chip cooling systems require the following components:

• A cooling liquid, which is typically a dielectric fluid engineered for direct-to-chip cooling
• A plate that the liquid can pass through
• A circulator that moves the liquid
• A thermal interface material, which helps conduct heat from the source to the cold plate

Advantages

Direct-to-chip cooling dissipates heat more rapidly and consumes less energy compared to traditional cooling methods, such as air-cooled server racks.

Targets specific heat sources:

• Unlike traditional air-cooling systems that blow cooled air across an entire array of equipment, much of the airflow passes over components like network interfaces and the peripheral parts of mainboards that are not significant heat sources, resulting in a considerable portion of the treated cooled air being effectively wasted.
• In contrast, direct-to-chip cooling systems can target specific heat-generating components. This enables them to dissipate more heat with less energy, benefiting both sustainability (by reducing data center carbon output through lower energy consumption) and reliability. The better systems are at dissipating heat, the lower the risk of IT failures due to overheating.

Challenges

• Cost: Chilled air systems are almost always less costly to implement and maintain than direct-to-chip systems, which require more expensive and complex components.
• Leakage risk: Fluids within direct-to-chip systems can leak. Although these fluids are non-conductive and generally do not pose a major threat to IT equipment, leaks can cause cooling systems to fail, potentially leading to overheating. Data centers worried about leakage risks may opt for immersion cooling, which is less prone to leakage because the fluid is not contained in small, fragile plates.
• Environmental impact: The fluids that flow through direct-to-chip systems are often not environmentally friendly, which means there are pollution risks associated with these systems.
• Uncooled components: Because direct-to-chip systems only cool major heat sources, there is a risk that other parts of a server, such as hard disks, could overheat. This is another reason why some data center operators opt for immersion cooling, which cools all parts of a system.

WGI’s MEP Expertise

Our MEP experts at WGI remain dedicated to delivering cutting-edge solutions that optimize performance and efficiency in data centers. Leveraging the latest advancements in cooling technologies, we provide sustainable, reliable, and cost-effective results tailored to meet your specific needs.

As we conclude our “” series, we invite you to explore the previous installments to gain a comprehensive understanding of the next-generation solutions available.

• Cooling Down AI & Data Centers Blog Series:
  • Part 1: Introduction to Cooling Systems
  • Part 2: Introduction to Liquid Cooling Systems

Contact Us

Innovate with confidence – Contact us today and explore how our services can elevate your data center!