Microsoft progresses with a modern way to cold the microchips, which he thinks can lead to more energy -saving data centers in the future. This is a method called micro -excavations, which includes liquid cooler flowing directly into silicon.
After laboratory tests, Microsoft stated that this strategy can remove heat up to three times better than frosty tiles currently used in data centers. This week, the company announced that it was able to develop a micro -ranking cooling system for a server with basic services for a simulated meeting of Microsoft teams.
If they manage to find the same success outside the laboratory, micro -fiddles can reduce the amount of energy needed to cold the data center. It can also lead to stronger systems that current cooling systems would have difficulty maintaining overheating. But there are still many factors that can affect how this modern technology affects is ultimately in the real world.
This can lead to stronger systems that current cooling systems will have difficulty overheating
Compared to Yore data centers, the next generation is built to train and run modern AI models that have stronger systems. These GPU not only consume a lot of energy, but also very heated. Keeping them in the frosty is a challenge that not only affects performance, but also causes data centers to consume more energy.
Usually the data center can exploit fans to pass cold air on the system. A more advanced technology, which Microsoft uses for greater power chips, includes frosty plates made of copper with liquid flowing through them. Put this plate on the chip, which pushes the heat.
When micro -flow cooling, the liquid flows through the channels with metatured on the back of the system. The trick is to make sure that the channels, with the width of human hair, are deep enough to prevent clogging, but not so deep that the chip becomes more likely. Microsoft claims that he used artificial intelligence to find out where to direct the coolant on the chip to cold it best. Extracted projects are also inspired by nature – for example, imitating the patterns of veins on the leaves – which have already shown how practical they are in the distribution of water and resources. By using the micro -ranges, Microsoft documented a 65 % reduction in the maximum enhance in Silicon GPU temperature.
The advantage of the micro -ranges is that it brings fluid straight to the system, eliminating the need for protective layers of materials between the system and the cooling fluid when frosty tiles are used. Each layer, like a blanket, holds in a certain heat, so the coolant must remain cooler to work well on frosty tiles. Cool liquid flows into the plate; The heated fluid flows and must be cooled again. In the case of micro -ranges, the coolant does not have to be chilled to low temperature, saving energy.
Microfluidics can also enable the data center for more effective support for demand peaks. The connections of the bands usually start every hour or half an hour, Microsoft gives as an example. To deal with these demands, they may have to install more servers to have sufficient capacity at hand, even if they are not used at all times. An alternative would be allowing existing servers to work challenging, called a fraudulent – but it can lead to overheating and damage to the system. Micro -flow cooling, because it is more effective, can allow more overcounts without the same melting risk.
Theoretically, if servers can work harder than now without melting chips, the data center may not require so many of them. And by minimizing the risk of overheating, micro -fate can also allow closer servers in one data center. This can reduce costs, literal and environmental, you need to build additional amenities.
All these benefits can be crucial for modern generation microchips, which are expected to become so powerful that frosty boards may not fall. Microsoft claims that micro -fate can also include 3D chip architecture. 3D chips would be even stronger than today’s one -half -flexible designs, but heat was an obstacle to make it happen. However, in the case of micro -ranges, it is possible to flow through the cooling fluid through the system.
Performance can also be a double -edged sword
Microsoft has no schedule when all this can happen. After more laboratory tests, there is a challenge for determining how to introduce changes in equipment and supply chain needed to enable micro -fantasies – for example, at what moment of the production process will be carved in the systems? Fortunately, they can exploit the same type of cooling fluid, a mixture of water and propylene glycol, used today in frosty tiles.
Other researchers have also been studying micro -fate for years. For example, HP received $ 3.25 million in Financing From the Energy Department last year to developing your own micro -flow technology. “All these things are good to see, we are happy to see them and where we can participate to move things faster,” says Husam Alissa, system technology director in the field of cloud operations and innovation at Microsoft.
Microsoft claims that “he hopes to pave the way for more efficient and sustainable new generation systems throughout the industry” in the last Blog post advertising your progress in the micro -ranges. Energy efficiency is crucial if the company wants to act more balanced. Like other technology companies, Microsoft carbon dioxide emissions grew when it was based on generative artificial intelligence. But performance can also be a double -edged sword. Because something becomes more effective and inexpensive to exploit, people tend to exploit much more, which can ultimately lead to an even greater environmental trace. This phenomenon called the Jevons paradox, which even the general director of Microsoft Nadella sata commented As a power that drives a greater reception of artificial intelligence.