Is there any way to use up the water of Hangzhou's West Lake?
This is the West Lake of Hangzhou, surrounded by mountains on three sides, with an area of about 6It is 39 square kilometers and is about 32 km, about 2 kilometers wide from east to west8 kilometers with a water body capacity of about 14.29 million cubic meters.
This amount of water can fill 3,810 water cubes, can fill 25 billion bottles of Nongfu Spring, and if you drink 4 bottles a day, you can give 214041 people from birth to 80 years old.
The West Lake water can fill my tears, but it's not enough for Google for 7 months.
Some time ago, Google released its 2023 environmental report.
An outrageous data in it immediately aroused everyone's attention.
Last year, Google spent 5.6 billion gallons of water, which translates to 25.45 million cubic meters, which is about enough to fill one and a half West Lake.
Of course, with so much water, it's definitely not going to do something like "Google Ice Spring" and "Microsoft Mountain Spring is a little salty".
I flipped through the report and found that the culprit pointed to the data center.
Most of this water is used to dissipate heat from the data center.
And it's not just Google that stands out, but also tech giants such as Microsoft and Amazon, all of which are big water consumers.
Yes, the heat dissipation of the data center is really too water-intensive...
In fact, in the early years, many data centers did not rely on water for heat dissipation, but directly used electricity to blow air conditioners on servers.
Take ourselves as an example, for example, in 2021, the total power consumption of data centers in the country will be as high as 216.6 billion kWh, accounting for about 2% of the total power consumption in the country6% 。
It may seem like a lot, but if it can be worth the money, it would be nice to use it all for data processing and storage.
But the problem is that there is a lot of electricity used to do "useless work".
At present, the biggest "useless work" is heat dissipation.
After all, we haven't pointed out the superconducting skill tree yet, and the data center is open all year round, and the heat generation is huge.
In order to make the instrument work normally in a suitable environment, it is necessary to rattle a large air conditioner and a large electric fan, which naturally consumes electricity.
According to statistics, 6% of the cost of the data center is spent on electricity, and more than 4% of the electricity bill has to be spent on heat dissipation.
So in order to save this money on heat dissipation, the best way is to prostitute as much as possible.
As a result, we see that various companies are racking their brains.
The first thing that people think of is "where to stay where it's cool".
In 2014, Tencent set aside Guizhou to dig out a 470,000-square-meter "Qixing Cave" to be used as a data center.
In 2017, Huawei built a cloud tun in Gui'an New Area to be used as a HUAWEI CLOUD data center.
In 2018, Apple's iCloud data also became "Guizhou on the Cloud".
Everyone went to Guizhou collectively, on the one hand, Guizhou has cheap land prices, clean air, abundant hydropower resources, cheap electricity bills, and many related preferential policies.
On the other hand, the environment of Guizhou is cool and stable enough to be like spring all year round, and the temperature difference between the external environment and the data center itself is large enough, which is more conducive to heat dissipation.
Compared with these domestic companies running to Guizhou, foreign manufacturers are more ruthless, Facebook as early as 2013, in a small town outside the Arctic Circle, set up its own data center.
In addition to moving to cooler places, the data centers themselves are constantly experimenting with new ways to cool their heat.
In this evolutionary process, the heat dissipation of data centers has also shifted from being very power-intensive to more water-intensive.
As mentioned earlier, the air conditioning cooling system has been gradually eliminated in recent years because it requires a lot of electricity to cool the condensate.
Is there a way to dissipate heat without using electricity, and at the same time it can be used for nothing?
That's right, water.
Compared with traditional air conditioners, evaporative coolers only rely on external water to cool down, and do not need electricity to cool the condensate, so they can greatly reduce power consumption.
But at the cost of a skyrocketing water consumption.
However, compared to electricity, water is always the cheaper one, so everyone generally starts to choose evaporative coolers.
In addition to evaporative coolers, in order to improve heat dissipation efficiency, some companies directly use liquid cooling for data centers.
When it comes to liquid cooling, netizens are not sleepy, after all, the 4090 water cooling with RGB light strip at home is not a dry meal.
But in the data center, water cooling is even more exaggerated.
It looks a bit like a Beijing root server.
For example, in Alibaba's data center in Qiandao Lake, although the equipment is soaked in a special coolant for rapid heat dissipation, these liquids still need to rely on a large amount of lake water to cool down.
Heat dissipation is carried out with the deep water source of Qiandao Lake.
In addition to Alibaba Cloud, there are also experts in the power saving of data centers. The most amazing is Microsoft, in 2018, they threw the "Northern Islands" server to the bottom of the Atlantic Ocean, using the tides of sea water for water cooling.
And it's also easy to build, and it will require less power backup than before.
In the past, the air conditioner dissipated heat, and the power backup also needed to drive the air conditioner, but after replacing it with liquid cooling, there is no need to worry about this part. This method is also convenient in places where water resources are relatively sufficient, saving electricity and trouble.
But in this world, water resources are never balanced. If it is still like this in some places where there is a shortage of water, isn't it a lack of great virtue?
Although the evaporated water vapor eventually returns to the atmospheric circulation, it takes a lot of time for the water vapor to flow back into the local water.
This point is well protected in our country, and in Inner Mongolia, where water resources are scarce, it is decided to name names:
Big data enterprises in the jurisdiction are prohibited from using groundwater for cooling and cooling. ”
However, in some water-scarce areas of the United States, the water consumption of this server has caused a lot of problems.
As we all know, the United States has been suffering from drought in recent years, and then several large factories are frantically building new data centers, many of which are built in Arizona.
Why did you choose this place to build this data center, this is not a blind choice.
First of all, this thing is next to California, which is engaged in technology, and the industrial electricity bill is much cheaper than California.
In addition, the land in the desert area is not expensive, and you can save a lot of money by building a data center here.
Moreover, in order to allow these data centers to settle down, they will also make a lot of preferential policies in taxes, and sometimes they will sign agreements to give priority to ensuring water quality**.
Everything was fine, and although Arizona was very dry, the Colorado River flowed through it, and it was enough to make do with the original groundwater.
But the problem is that recently, the Colorado River has also begun to run out of water, and the landlord's family has no surplus food.
As soon as several factors add up, things get troublesome.
For example, when Google built a data center in Mesa, Arizona, it signed an agreement with the local **, and Mesa wanted to "prioritize" to ensure that Google had enough water to dissipate heat every day.
This has affected the local ecosystem and even the water supply of human life.
Similarly, the state of Arizona**, which was forced to stop urban infrastructure because of its water supply to the data center, was also taught a lesson by the U.S. federal government**: "Use less river water".
In The Dalles, the annual water consumption of Google's three data centers has reached nearly one-third of the city's total annual water consumption, and it has been chased and bitten by the local ** all the way.
As for the establishment of large-scale recycling equipment, the method of recycling water vapor energy has not been mentioned, but due to the high construction cost, the maintenance cost is far away, and the cost of electricity and other reasons, it has basically been shelved.
Fortunately, at this time, more and more people are paying attention to these manufacturers who use water indiscriminately.
Various large manufacturers also had to soften and began to optimize their own heat dissipation equipment.
For example, Google promised that by 2030, whether it is diverting water from east to west or investing in seawater purification facilities, no matter how much water it uses, it will have to pay 120% of the water in return.
Microsoft, for its part, has pledged to reduce the water consumption of evaporative cooling systems in data centers worldwide by 95% by 2024 and achieve "water neutrality" by 2030
However, progress towards their commitment has been a real concern, and according to Google's own report, it has only achieved 6% of its replenishment so far.
Why, maybe ChatGPT is to blame.
With the explosion of generative large models, more and more water resources are being consumed. According to Shaolei Ren, an associate professor at the University of California, Google uses 20 percent more water than last year, and coincidentally, Google's computing power has also increased by 20 percent this year.
So they think proactively, make bold assumptions, and analyze calmly – it's because of the AI arms race that data centers consume more water. According to a study by the University of California, Riverside, ChatGPT uses a lot of water for large language models.
According to their calculations, OpenAI uses 700,000 liters of water to train GPT-3 alone, not to mention GPT-4, which has a larger amount of data.
And in addition to the training stage, the follow-up use is also particularly water-intensive, and a casual chat with GPT for thirty or fifty sentences will consume a bottle of Nongfu Spring water.
At present, the AI boom continues, everyone's AI competition will only become more and more fierce, and the role of data centers will only become more and more important.
Whether it is domestic or foreign science and technology manufacturers, when they are under construction, they also pay attention to water resources and other related issues.
Otherwise, before AI takes away human jobs, the water that some people rely on for survival may be snatched away by AI first.
Written by: Feathers fall silently
Editor: Huan Yeon.
*The content represents the author's views only.
It does not represent the position of the Institute of Physics of the Chinese Academy of Sciences.
If you need it, please contact the original***
*: Negative. Editor: Nilo.