The future of data centers: the convergence of artificial intelligence and liquid cooling

The rapid rise of generative artificial intelligence (AI) highlights the breakneck pace at which businesses are adopting AI. According to a recent Accenture report, 98% of business leaders say artificial intelligence will play an important role in their strategy over the next three to five years. McKinsey analysts find that nearly 65% ​​of enterprises plan to increase investment in artificial intelligence in the next three years

NVIDIA, AMD and Intel are launching new technologies designed for generative AI and high-performance computing (HPC) Chips, this momentum has just begun. Public cloud providers and emerging chip companies are also competing. IDC analysts predict that global spending on artificial intelligence software, hardware and services will reach $300 billion, exceeding the $154 billion expected this year

However, there are still challenges in scaling artificial intelligence, the most important of which involve Challenges with the data center infrastructure required to support these workloads.

Data centers are becoming more and more "hot"

GPU is the most common chip in artificial intelligence and machine learning, which can accelerate the computing process of artificial intelligence applications. For example, NVIDIA's H100 GPU has 80 billion transistors, so it generates a lot of heat and requires efficient cooling. Traditionally, configurations reaching 10 kilowatts in a single data center rack have been considered high density. But air cooling is still an effective way to cool these servers. Although the Uptime Institute found that few data centers have racks exceeding 30 kilowatts, extreme densities are emerging. The commoditization of high-performance computing and the rise of generative artificial intelligence are increasing power demands and overtaxing traditional air cooling methods.

For example, NVIDIA’s latest GPU’s maximum power consumption is 160% higher than the previous generation chip. Rack configurations can easily exceed the 40kW range, which is difficult to manage with traditional air cooling methods. Today's data centers must continue to evolve to effectively manage these increased heat loads

Cooling Technologies Are Increasingly Important

Fortunately, we have a variety of liquid cooling technologies that can meet this challenge, including Backdoor hot-swap and direct-to-chip technologies are becoming increasingly popular. There are also different types of emerging immersion cooling technology, which essentially involves immersing IT components in a container filled with liquid coolant. Although immersion cooling is still in its early adoption stages, analysts predict that the technology It will become mainstream in the next four years, with the market size growing from US$251 million in 2021 to more than US$1.6 billion in 2027. This will significantly impact data center infrastructure needs, and business leaders must know whether their data center operators are willing to make the necessary investments in the short term to support this shift.

Advantages and Disadvantages of Liquid Cooling

Liquids are 1,000 times more efficient as heat conductors than air and require less infrastructure. Air cooling systems require complex refrigeration equipment, including coolers, air pumps, cables, humidity control and filtration systems, and redundant backup systems to ensure that servers do not lose cooling during a power outage

In contrast, Liquid cooling systems are relatively simple, but implementing them in current data center infrastructure can present significant challenges, including upfront investment and complexity. Setting up a liquid cooling system can be complicated and may require specialized maintenance. Additionally, server designs may need to be adjusted, adopting an immersion approach may void the OEM warranty, and cooling system leaks may cause equipment damage and downtime. Data center operators must also take into account new regulations and environmental standards involved in using liquid cooling systems. That said, liquid or immersion cooling systems do not require as much backup or special floor or aisle sealing strategies. . The overall impact on energy consumption and costs can be significant. Results of a recent study found that implementing liquid cooling can reduce facility power by nearly 20% and total data center power by more than 10%. Total Usage Effectiveness (TUE), a new metric designed to compare the efficiency of liquid cooling to air cooling in high-performance computing environments, shows that liquid cooling improves energy efficiency by more than 15%.

Transitioning to liquid cooling has other sustainable benefits. Liquid cooling systems require less water than air cooling systems. Retrofitting data centers can employ new ways of thinking to shrink their physical and carbon footprints. Thermal reuse strategies can provide energy to surrounding businesses and communities. The possibilities are exciting and could be as transformative as generative AI itself.

What to Know Now

For most enterprises, transitioning to an on-premises data center may be too complex and expensive. On the other hand, much of today’s public cloud infrastructure is not built to run large-scale AI applications, and the rising cost of hosting high-volume workloads in the cloud is prompting many organizations to look for other options

Given these challenges and opportunities, colocation data center providers with infrastructure experience handling myriad customer use cases may provide the best solution for many enterprises. Leaders in this space can provide expertise and support to guide organizations through their transformation. We have also developed key relationships with a number of hardware OEMs and liquid cooling suppliers that will drive data center growth, providing diverse options to meet our customers' unique needs.

Organizations now need to know whether their data center operators are already planning and, perhaps more importantly, have the physical capacity available or the technology needed to fit in to enable the development of next-generation data centers. . Data centers already face the complex challenge of moving workloads to the best servers for their requirements. As the demands of artificial intelligence and high-performance computing workloads continue to increase, these obstacles will certainly be compounded by the additional challenge of adding fundamentally different cooling systems.

Data center operators that are currently investing in these strategies will be well-positioned to help their customers proactively address these challenges. Artificial intelligence is changing everything, including data centers. Now is the time to start this conversation

The above is the detailed content of The future of data centers: the convergence of artificial intelligence and liquid cooling. For more information, please follow other related articles on the PHP Chinese website!