With the increasing energy consumption of data centers, will gallium nitride be an effective savior?

January 8, 2024

The digital economy is undergoing a significant transformation driven by two major trends: real-time data analysis and the rapid development of Generative Artificial Intelligence (Generative AI). Currently, a fierce competition in Generative AI is underway, with tech giants such as Amazon, Google, and Microsoft heavily investing in the research and innovation of Generative AI.


According to Bloomberg Intelligence predictions, the Generative AI market is expected to grow at a rate of 42% annually, increasing from a $40 billion market value in 2022 to $13 trillion within a decade. With the flourishing development of AI, the demand for power and computation in data centers is growing proportionally. The surge in power consumption not only puts pressure on operational efficiency but also becomes a hindrance to data centers achieving their net-zero emission goals(sources from medcom.com.pl).


The power conversion and distribution technologies currently used in data centers are struggling to meet the computational demands from cloud computing and machine learning. Faced with the larger energy requirements of Generative AI applications, data center operators are urgently seeking innovative power solutions.


In recent years, Gallium Nitride (GaN) power semiconductors have emerged as a key technology for optimizing energy efficiency in data centers. The following article will further explore the challenges Generative AI poses to data centers, the advantages of GaN power devices, and analyze the flaws in the commonly used Power Usage Effectiveness (PUE) metric. These insights all point to one conclusion: the application of GaN power components in data center power systems will accelerate.


Impact of Generative AI on Data Centers Intensifies

Faced with the explosion of data, data center infrastructures planned for traditional workloads are under tremendous pressure. Global data volume doubles every two years, and servers processing and storing this massive data require a significant amount of energy and water resources to operate.


According to McKinsey's predictions, within the next five years, data centers in the United States alone will add 39 GW of electricity consumption, equivalent to the annual electricity consumption of 32 million households. While Generative AI brings disruptive innovation and business opportunities, the unprecedented energy demand it brings is undeniable.


A research article points out that training the latest generation of Generative AI models requires ten to a hundred times more energy than previous models. The energy consumption of Generative AI applications primarily comes from two aspects: the energy required to train large language models (LLMs), which constitute the core of Generative AI systems, and the energy consumption during the runtime of large language models.


The power consumption of Generative AI is unbelievably high. Assuming each Google search consumes the equivalent of the energy required for a 100W light bulb to glow for 11 seconds, the popular Generative AI tool ChatGPT consumes 50-100 times more energy to answer a single question than a Google search. Faced with the sharp increase in power demand, upgrading data center infrastructure is imminent, and data center operators explicitly state that the estimated cost of upgrading equipment in the next four years will reach up to $1 trillion.


Although emerging technologies such as immersive cooling, artificial intelligence optimization solutions, and waste heat reuse have appeared, these solutions cannot fundamentally solve the problem. Due to the low efficiency of power components based on silicon development, today's data centers heavily rely on cooling systems to maintain a safe operating temperature. Therefore, a power solution with both energy efficiency, size advantages, and output power is urgently needed in current data centers.


Gallium Nitride Exhibits Excellent Efficiency and Performance

Compared to traditional power supplies, power supplies designed based on Gallium Nitride power components can achieve outstanding performance and efficiency, making them an ideal choice for data centers as Generative AI applications continue to evolve.


With its faster switching speed and outstanding Figures of Merit (FOM), gallium nitride (GaN) realizes advantages in power supply design, delivering energy efficiency performance surpassing titanium levels and significantly increasing power density. GaN transistors bring higher efficiency levels to power devices, and increased efficiency means reduced energy loss, thereby mitigating overheating issues in equipment.


For example, in a typical data center architecture, power supplies designed with GaN can annually increase revenue by $3 million for every 10 racks, reduce carbon dioxide emissions by 100 metric tons, and save $13,000 in operational expenses. As the power demand for Generative AI continues to rise, and the power density per rack increases to 2-3 times, the advantages of GaN become even more significant.


You might wonder why data center operators haven't immediately adopted this technology, given its apparent benefits. The answer lies in a frequently overlooked blind spot in the widely used Power Usage Effectiveness (PUE) metric, which we refer to as the "PUE Loophole." This is a major factor limiting the widespread adoption of GaN technology.


What is the PUE Loophole?

PUE represents the ratio of IT equipment load power consumption to the total power consumption, commonly used as a metric to evaluate data center energy efficiency. Monitoring and continuously improving PUE are crucial for optimizing data center operations, as lowering PUE indicates a reduction in electricity consumption, leading to cost and environmental impact reductions.


However, PUE only calculates the efficiency of power transmitted to servers, neglecting the energy losses during power conversion within servers. This is the PUE Loophole, a blind spot in evaluating overall operational efficiency when using PUE as a metric.


This blind spot results in many servers still using AC-DC converters with efficiency as low as 90%, or even lower. This means that 10% of energy is wasted during each AC-DC energy conversion, increasing electricity costs and carbon emissions. The wasted energy is also released in the form of heat, increasing the data center's demand for additional cooling systems, which are another power-consuming component(quotes from medcom).


GaN power semiconductors will revolutionize this challenge by providing data centers with an immediately effective and cost-efficient solution to address the PUE Loophole, saving substantial power requirements.


AC-DC converters using GaN power components can achieve efficiencies of up to 96% or higher, resulting in a 50% reduction in power consumption. In terms of the average power consumption of data centers, this is equivalent to saving 37 billion kilowatt-hours of electricity, enough to meet the power needs of 40 large-scale data centers for a year


Urban Splatter

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts
February 27, 2024
Cracks and Coatings: A Comprehensive Guide to Stucco Inspection

As a stucco-siding homeowner, you know that your home’s exterior needs regular maintenance and checking. Stucco, a popular siding material with an attractive and durable finish, can also develop cracks, discoloration, and other problems over time. In addition, stucco might be vulnerable to damages caused by weather patterns, moisture content in the air, or settling […]

Read More
February 27, 2024
Enhance Your Curb Appeal With These 3 Concrete Driveway Design Ideas

Enhancing your home’s curb appeal involves more than just the right landscaping and a fresh coat of exterior paint; it requires attention to every detail, including those often overlooked aspects such as the driveway. This crucial element of your property’s exterior can significantly influence the first impression it makes. With thoughtful design, high-quality materials, and […]

Read More
February 27, 2024
6 Weed Management Tips For A Pristine Lawn

Weeds – you see them popping up on your lawn and instantly dread the takeover you know is coming. They start as just a few interlopers, but before you know it, there's a full-on invasion covering up your formerly green grass. It's enough to make you want to hang up your gardening gloves and wave […]

Read More
Welcome to Urban Splatter, the blog about eccentric luxury real estate and celebrity houses for the inquisitive fans interested in lifestyle and design. Also find the latest architecture, construction, home improvement and travel posts.
© 2022 UrbanSplatter.com, All Rights Reserved.
linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram