Therefore,
Ai: new means reduce its new: Nevertheless. Consequently,
Ai: new means reduce its:
Faced with frantic growth
AI seeks the means to reduce its energy needs – Ai: new means reduce its – Ai: new means reduce its new
Liquid cooling. Meanwhile, optimized chips, intelligent algorithms: the sector multiplies innovations to reduce its consumption.
Posted today at 3:11 p.m.
AI data. However, centers are consumering more energy that raises many questions for the future.
AFP
Thanks to new cooling techniques. Therefore. However, more efficient fleas or the evolution of programming, the artificial intelligence sector (IA) seeks to limit its consumption energyin a context of frantic growth.
Infrastructure is based on data centerswhich should weigh. Similarly. For example, according to the International Energy Agency (AIE), around 3% of ai: new means reduce its the world’s electricity needs by 2030, double the current proportion.
On Tuesday. Similarly. Donald Trump is scheduled to go to Pennsylvania to announce, according to several media, some $ 70 billion in this state for AI and ai: new means reduce its new energy infrastructure.
McKinsey’s office talks about “race” to build enough sites to “respond to the massive ACA acceleration”. Therefore, while warning that a shortage is looming.
Cooling solutions – Ai: new means reduce its – Ai: new means reduce its new
“There are several ways to deal with this problem. ” said Mosharaf Chowdhury, professor at the University of Michigan. However, “You can increase energy sources. ” a way in which IA heavyweights are also engaged, “or reduce the demand” of electricity to equivalent capacity, he says.
For the academic. “malignant” solutions can be found at all levels of the ai: new means reduce its AI chain, from physical equipment to algorithms.
According to Gareth Williams. from the Arup consulting firm. today, the energy necessary to maintain a data center ai: new means reduce its new represents 10% of what the servers themselves consume, against 100% 20 years ago.
This reduction is to be put in credit. among other things. to generalize liquid cooling, in place of conventional ventilation, which goes so far as to circulate fluids directly in the servers.
“All who are using it now,” observes Gareth Williams, “and in the products they offer, it has become essential.”
The new chips of the King of Nvidia AI have multiplied by more. than 100 the consumption of a server cabinet compared to what it represented 20 years ago.
As a ai: new means reduce its result. the liquid can rise to much higher temperatures than before. according to Gareth Williams, but this makes it easier to cool in contact with the ai: new means reduce its new outside air, due to the temperature difference.
Amazon unveiled. in early July. a new liquid cooling system called Irhx, which can be installed in a data center without having been integrated into the initial architecture.
Progress on fleas
Another evolution. the data centers are now equipped with sensors. used by the AI to control the temperature no longer on the scale of a site but by “micro-zones” and “optimize the consumption of water and electricity” in advance, according to Pankaj Sachdeva, of the McKinsey office.
The Mosharaf Chowdhury laboratory has developed algorithms to precisely assess the quantity of. electricity necessary for each chip ai: new means reduce its to operate with. at the key, a gain of 20 to 30%.
Progress has also ai: new means reduce its new been recorded on the microprocessors themselves. “Each generation of chip is more efficient in terms of energy” than the previous one, recalls Pankaj Sachdeva.
Yi Ding’s team. professor at the University of Purdue. has shown that we could extend the life of the most efficient fleas for AI, GPUs (Graphics Processing Unit) or graphics cards, “without lessening performance,” she said to AFP.
“But it is difficult to convince semiconductor manufacturers to make less money” by. encouraging consumers to use the same equipment longer. slides the academic.
Ia générative
The match is also played in terms of programming and training of large generative AI models.
In January. the Chinese Deepseek thus presented its generative R1 model with performance similar ai: new means reduce its new to those of the great American actors. although developed with less powerful GPUs.
The engineers of the start-up succeeded in particular by programming more precisely the graphics cards. They have also almost jumped a stage of training of the model, deemed essential so far.
However. despite these technological breakthroughs. “we will not be able to reduce the total energy consumption, because of the Jevons paradox”, predicts Yi Ding.
The British economist William Stanley Jevons (1835-1882) proclaims as well as more effective. use of a limited resource makes demand increase the demand because its cost decreases.
“Energy consumption will continue to rise. ” warns Yi Ding, despite all the efforts to limit ai: new means reduce its it, “but perhaps less quickly”.
“Latest news”
Do you want to stay at the top of the info? “24 hours” offers you two meetings a day. not to miss anything of what is happening in your canton, in Switzerland or in the world.
Other newsletters
AFP
Did you find an error? Please report it to us.
Further reading: USA -The values to follow at Wall Street (updated) – 07/15/2025 at 13:15 – plane to the tax reduction – “Contempt is not drunk”, the winegrowers have carried out actions in Lidl stores – F-35 case: For Oswald Sigg, the Federal Council dropped Viola Amherd – Donald Trump accuses the Fed boss to waste public money.
Further reading: It’s official, here is how much your gas bill will increase – Artificial intelligence: Duolingo revolts users – Scam alert: Discover these formidable phone calls to which to respond could cost thousands of euros in just a few seconds – The values to follow in Paris and in Europe – 07/14/2025 at 06:54 – The demand for refuge values propels gold to a three -week summit; Money is silent almost 14 years old.
