Artificial intelligence (AI) is a subject of great debate when it comes to ethics, but one area people might not think about is its carbon footprint.
A study released last year by MIT Technology Review found that training a “regular” AI using a single high-performance graphics card has the same carbon footprint as a flight across the United States. Training a more sophisticated AI was even worse, pumping five times more CO2 into the atmosphere than the entire life cycle of an American car, including its manufacturing.
Whether it’s the latest AI or machine learning algorithm that’s active on a system, a new 5G network deployed at a commercial building or people streaming the latest Twitch gaming video, people generate and consume a lot of data. All that data must be captured, stored, analyzed and sent back out, which requires significant amounts of processing power. How can the tech industry deal with the increasing environmental cost of AI and its supporting systems while still providing the same service consumers demand?
Tallying The Cost Of AI
Just think of all the technology that surrounds you today. Now, think about all the data it generates, sends, consumes and analyzes. It’s hard for our human brains to comprehend the sheer amount of it floating around, but it’s forecast to be more than 33 zettabytes (ZB) right now. By 2025, we’ll top 175 ZB worth of data being stored globally. (That’s 175 trillion gigabytes, or 175 plus 12 zeros.)
Keeping data centers cool enough to allow AI and other processing devices to work efficiently is going to get harder every year. Some technology has gotten better at reducing the amount of heat they produce, and we’re virtualizing a lot more of it, but it’s not enough. The amount of processing and edge technology that networks need, and that consumers are demanding, make those gains moot. Some analysts have noted that analytics and machine learning are driving up the heat density of data centers again, to the point that they’re having the same problem as they did a decade ago.
Start Planning Today
The problem of cooling data centers, plus the knowledge that their carbon footprint is expanding and becoming more of a global problem, is leading the technology industry to think more strategically about cooling. Global awareness of climate change and how the technology industry can help is just one of the reasons they’re investing in it. Many AI and data center customers are also starting to demand their vendors and suppliers take responsibility for their part in climate change and are looking for vendors that are demonstrating it.
Cloud providers like Microsoft have announced their intentions to be carbon neutral by 2030 by developing tools to help customers understand the carbon footprint of their Azure deployments. Amazon Web Services (AWS) has committed to using more renewable resources, such as wind and solar energy farms. Even independent colocation facilities are getting in on sustainability. Switch, a colocation provider with facilities located primarily in the U.S., has been running on green energy since 2016. The company has even constructed a solar power facility in Las Vegas, which it uses to power its facilities and remove the equivalent of 50,000 cars off the road.
Explore Alternative Cooling Methods
Liquid cooling in data centers has become more popular as it’s more efficient than air cooling and can handle the higher temperatures cutting-edge processing technology emit. Jayarama Shenoy, vice president of technology at Hyve Solutions, explained to TechTarget that “Most data centers use three main types of liquid cooling: closed-loop liquid cooling, water-based cooling and liquid immersion cooling.”
Liquid cooling is more efficient at cooling because it can be installed closer to the device that needs it. That’s got the added advantage of lowering electricity costs because they use less electricity overall. It uses fewer kilowatt-hours to cool high-level processing devices, lowering overall cooling costs for the data center and reducing the carbon footprint.
Another exciting development in data center cooling is facilities that use their local geography for natural cooling. Locating facilities in colder climates such as Scandinavia reduces the amount of energy needed to cool them. In many industries, Iceland uses geothermal cooling technology, and some are using it to cool their data centers. Verne Global, a high-performance data center in Iceland, has been using it for over a decade, offering 100% free cooling year-round for its customers.
Learning more about the environmental impact of AI and other emerging technologies is the responsibility of the technology industry. We’ve all got to understand how we’re affecting more than our target customers and learn about ways we can minimize our impact. It’s one thing to say we’re committed to the environment and donate to climate change causes. It’s quite another to take proactive steps to ensure our technology keeps both people and the planet in mind. We’re all in this together.
To read the original article, click here.