A.I. usage fuels spike in Microsoft’s water consumption

But Microsoft-backed OpenAI’s technology requires vast amounts of water, which is pumped from the Raccoon and Des Moines river watersheds in central Iowa, to cool a powerful supercomputer as it helps the church How artificial intelligence systems can imitate human writing.

as they compete to take advantage of The craze for generative AILeading technology developers including Microsoft, OpenAI and Google acknowledge that growing demand for their artificial intelligence tools comes with significant costs, from expensive semiconductors to increased water consumption.

But they tend to be tight-lipped about specific details. Few people in Iowa knew it was the birthplace of OpenAI’s state-of-the-art large-scale language model, GPT-4, until a Microsoft executive said in a presentation that it was “literally built next to a cornfield west of Des Moines.” .

Building large language models requires analyzing patterns in large amounts of human-written text. All these calculations consume a lot of power and generate a lot of heat. To stay cool on hot days, data centers need to pump water—usually into cooling towers outside warehouse-sized buildings.

In its Latest environmental reportMicrosoft revealed that its global water use jumped 34% from 2021 to 2022 (to nearly 1.7 billion gallons, the equivalent of more than 2,500 Olympic-sized swimming pools), which is comparable to previous years when outside researchers have linked to its AI research. This number has increased dramatically.

Shaolei Ren, a researcher at the University of California, Riverside, said that “it’s fair to say that much of the growth is due to artificial intelligence,” including “significant investments in generative AI and collaboration with OpenAI.” The environmental impact of generative AI products like ChatGPT.

In a paper to be published later this year, Ren’s team estimates that every time you ask ChatGPT a series of 5 to 50 prompts or questions, ChatGPT drinks 500 milliliters of water (nearly the amount in a 16-ounce water bottle). amount of water). This range varies based on server location and season. The estimate includes indirect water use that the company doesn’t measure, such as cooling the power plants that power the data centers.

“Most people don’t know the resource usage under the hood of ChatGPT,” Ren said. “We can’t help save resources if you don’t understand how they are being used.”

Google report Water consumption increased by 20% during the same period, which Ren Zhengfei also attributed largely to his artificial intelligence work. Google’s spike isn’t even – it remains steady in Oregon. attracted by water Public attention doubled outside Las Vegas. Iowa is thirsty, too, sending more drinking water to its Council Bluffs data center than anywhere else.

In response to questions from The Associated Press, Microsoft said in a statement this week that it was investing in research to measure the energy and carbon footprint of artificial intelligence “while working on how to make large systems more efficient in training and application.” .

“We will continue to monitor our emissions and accelerate progress while increasing the use of clean energy to power data centers, purchasing renewable energy and other efforts to achieve our sustainability goals of being carbon negative, water positive and zero waste by 2030. ” a company statement said.

OpenAI responded to those comments in a statement on Friday, saying it was “thinking hard” about how to best use computing power.

“We recognize that training large models can be energy and water intensive,” and are working to improve efficiency, the statement said.

Microsoft made its first $1 billion investment in San Francisco-based OpenAI in 2019, more than two years before the startup launched ChatGPT and sparked a global craze. Fascination with advances in artificial intelligence. As part of the deal, the software giant will provide the computing power needed to train artificial intelligence models.

To do at least some of that work, the two companies are looking to West Des Moines, Iowa, a city of 68,000 people where Microsoft has been building data centers for its cloud computing for more than a decade. Services provide support. Its fourth and fifth data centers will open there later this year.

“They’re building them as fast as they can,” said Steve Gaer, who was the city’s mayor when Microsoft arrived. Gale said the company was attracted by the city’s commitment to building public infrastructure and contributed an “amazing” amount of money through tax payments that supported the investment.

“But, you know, they were very secretive about what they were doing there,” he added.

Microsoft first said it would develop one of the world’s most powerful supercomputers for OpenAI in 2020, declining to disclose its location to The Associated Press at the time but describing it as a “single system” with more than 285,000 traditional semiconductor cores and 10,000 graphics processors – one Chips become critical Artificial Intelligence workloads.

Experts say that because of the large amounts of data that need to be transferred between computing cores, it makes sense to “pretrain” AI models in a single location.

It wasn’t until late May that Microsoft President Brad Smith revealed that the company had established an “advanced artificial intelligence supercomputing data center” in Iowa specifically for OpenAI to train its fourth-generation model. GPT-4. The model now powers advanced versions of ChatGPT and some of Microsoft’s own products, and has accelerated the debate about curbing the risks to society of artificial intelligence.

“It was made by these great engineers in California, but it was actually made in Iowa,” Smith said.

In some ways, West Des Moines is a relatively efficient place to train powerful artificial intelligence systems, especially compared to Microsoft’s data centers in Arizona, which consume more water to serve the same computing needs.

“So if you’re developing an AI model within Microsoft, you should schedule your training in Iowa rather than Arizona,” Ren said. “As far as training goes, there’s no difference. Whether it’s water consumption or energy consumption, there’s a big difference.”

For much of the year, Iowa’s weather is cool enough that Microsoft can use outdoor air to keep supercomputers running and heat out of the building. The company said in a public disclosure that it will only draw water when the temperature exceeds 29.3 degrees Celsius (about 85 degrees Fahrenheit).

That’s still a lot of water, especially in the summer. July 2022, one month ago OpenAI says it’s done According to the West Des Moines Water Works, Microsoft injected approximately 11.5 million gallons of water into its Iowa data center cluster during GPT-4 training. This accounts for approximately 6% of the total water consumption in the region, which also provides drinking water to the city’s residents.

In 2022, a West Des Moines Water Works document said the company and the city would “only consider Microsoft’s future data if the projects can “demonstrate and implement technologies that can significantly reduce peak water use from current levels.” Center Project”. Preserves water supply for residential and other commercial needs.

Microsoft said Thursday it was working directly with the waterworks to address its feedback. The company said in a written statement that the company has been a good partner and has been working with local officials to meet demand while reducing its water footprint.

—-

O’Brien reported from Providence, Rhode Island.