As artificial intelligence continues to redefine the computing landscape, the heat dissipation challenges posed by AI chips demand increasingly sophisticated cooling solutions. At SC24 in Atlanta, a panel of experts from Dell Technologies, NVIDIA, and Seguente shed light on the latest advancements in liquid cooling. From hybrid cooling systems to cutting-edge two-phase technologies, the discussion underscored the industry’s pursuit of efficient, scalable, and sustainable cooling solutions.
Liquid Cooling in the AI Era: Where Are We Today?
The panel opened by revisiting the journey of cooling systems, from air-cooled data centers to the resurgence of liquid cooling. AI systems, particularly those leveraging GPUs like NVIDIA’s Hopper and Blackwell architectures, are pushing the limits of thermal management. The panelists highlighted key cooling methods in use today:
- Air Cooling and Hybrid Systems: Dell emphasized the continued relevance of air-cooled systems, noting that many data centers remain reliant on this method. However, hybrid approaches, such as combining air cooling with rear-door heat exchangers, are gaining traction. These systems allow liquid cooling at the rack level while keeping water out of the server chassis, balancing efficiency and reliability.
- Direct Liquid Cooling (DLC): Single-phase direct-to-chip cooling is becoming the go-to solution for many hyperscalers. This method uses water or a water-glycol mixture to transfer heat via cold plates attached to CPUs and GPUs. Dell highlighted its ability to scale this approach to meet the demands of massive deployments, including racks with 3,000-watt GPUs.
- Two-Phase Cooling: Represented by Seguente, two-phase cooling introduces dielectric fluids with low boiling points. These fluids absorb heat through evaporation and condense back into liquid without the need for pumps, leveraging passive circulation. This innovation minimizes energy consumption and addresses concerns about leakage, as the fluids are non-conductive.
The Promise of Two-Phase Cooling
Dr. Luca Amalfi, CEO of Seguente, advocated for the transformative potential of two-phase cooling, describing it as an end-to-end solution that aligns with both performance and sustainability goals. Key advantages include:
- Energy Efficiency: Two-phase cooling leverages phase change (evaporation and condensation) to store and transfer heat, reducing the need for large-scale pumping and HVAC systems.
- Scalability: Seguente has designed its systems to integrate seamlessly into existing data centers, using the same cold plates and infrastructure as single-phase systems.
- Safety and Reliability: The use of dielectric fluids mitigates risks associated with leakage, ensuring that even in rare cases of failure, sensitive electronics remain unharmed.
- Material Flexibility: Two-phase cooling systems can utilize aluminum instead of copper, reducing costs, weight, and environmental impact.
Challenges and Industry Adoption
While the benefits of two-phase cooling are clear, the panelists acknowledged hurdles to widespread adoption. Tim Shedd of Dell pointed out the engineering complexities of scaling new technologies to tens of thousands of servers. Additionally, customers must weigh the total cost of ownership (TCO) and the readiness of their infrastructure for advanced cooling solutions.
NVIDIA’s Dion Harris emphasized the role of ecosystem collaboration in overcoming these challenges. NVIDIA actively participates in research initiatives like the DOE-sponsored Coolerchips project to explore innovative cooling methods. Harris noted that while direct liquid cooling is the standard for current systems, NVIDIA remains open to supporting diverse approaches, including two-phase cooling, to address future demands.
The Road Ahead: Innovation and Sustainability
Looking to the future, the panelists agreed that liquid cooling will play a critical role in enabling high-density computing. As workloads grow more complex and energy-intensive, the industry must balance performance with sustainability. Two-phase cooling, with its ability to capture and reuse heat, offers a promising path forward.
Dr. Amalfi concluded with a vision for integrating cooling and power innovations to achieve greater efficiency and environmental responsibility. “As we increase the power of racks, capturing heat at high temperatures for reuse will be key to sustainable data center operations,” he said.
Final Thoughts
Liquid cooling is no longer a niche solution—it is an essential enabler of AI’s rapid growth. From hybrid systems to advanced two-phase technologies, the industry is evolving to meet the demands of accelerated computing. While challenges remain, the collaborative efforts of companies like Dell, NVIDIA, and Seguente are paving the way for scalable, efficient, and sustainable data center architectures.
As the race for innovation continues, one thing is clear: the future of AI cooling is as dynamic as the technology it supports.
Watch the full video:
