By Bernie Malouin, CEO, JetCool
In the world of data centers, it’s a lesser-known fact that the most critical element isn’t the servers, chips, or racks, but the cooling systems that keep them operational. With greater frequency, traditional methods, predominantly air-based cooling, are proving inadequate for the intense heat generated by the latest generation chipsets. The economic and environmental cost of running an army of fans and air conditioners is becoming unsustainable.
Liquid cooling, specifically immersion cooling, offers lower energy consumption and improved computing power. However, its adoption in large-scale data centers and colocations has been limited, primarily due to concerns about fluid spills, insurance costs, maintenance, and the high cost and time required to change the data center’s architecture for horizontal tanks, particularly in industries where data integrity is paramount and every second of downtime can cost millions of dollars.
But there’s an alternative approach to liquid cooling that addresses these challenges. Single-phase direct-to-chip liquid cooling, particularly effective with “microjet impingement,” targets coolant directly at the hottest chip parts, managing high heat loads while boosting silicon performance. This method, which can efficiently cool chips over 2,000W, offers a significant improvement over the cooling capacity of traditional single- and two-phase immersion systems.
Direct-to-Chip Solutions Enable Higher Computing Power
Leveraging these cooling advancements not only effectively manages high heat loads but also boosts silicon performance. Their ability to cool chips over 2,000W greatly surpasses the capacity of traditional single-phase immersion cooling, which generally maxes out at 400W TDP, and two-phase immersion cooling at around 1,000W.
Building on the benefits of microjet impingement, the latest development in this field is the creation of a self-contained liquid cooling system for servers. This system adapts microjet impingement to cool servers up to 1,200W within a compact 1U and 2U form factor with no external piping, coolant distribution units (CDUs), or dry coolers.
The financial and environmental benefits are notable. Upfront installation costs in data centers and colocations are 50 percent less compared to traditional liquid cooling in part because there’s no need for extensive plumbing or structural modifications. As chips evolve and facilities seek to replace legacy servers, self-contained liquid cooling (also called liquid-assisted air cooling) enables plug-and-play deployments, allowing organizations to implement the upgrades with minimal investment.
Microjet impingement also frees up increasingly limited rack space because it is small and streamlined, leaving more real estate for computing equipment. Moreover, these systems use water, moving away from environmentally harmful coolants used in two-phase systems. Energy conservation is also notable, with net savings of 100W per server. This approach, better for the planet and more cost-effective, is scalable and adaptable, fitting a range of computing environments.
These self-contained systems also present a low leakage risk, a significant issue for CTOs and decision-makers. The systems need minimal fluid, only 10 ounces in a 2U server, reducing the risk of infrastructure damage or operational downtime, ideal for the thermal challenges posed by high-powered chips.
Innovative Cooling for Next-Generation Chips
The self-contained nature of these liquid-assisted air-cooling systems, with their minimal fluid requirements and low leakage risks, addresses concerns that have previously deterred facilities from adopting liquid cooling. Suitable for various computing environments, from emerging AI startups to established data centers, these systems signify a shift towards more environmentally friendly and cost-effective cooling solutions.
Bernie Malouin is CEO of JetCool, Littleton, MA, which designs cooling modules that focus on decreasing energy consumption, water usage and carbon footprints.
Speak Your Mind