The "new infrastructure" came out, blowing a pool of spring water.
The setting of the top-level design and the continuous increase of policies mean that the corresponding industrial dividends are also arriving as scheduled.
Whether it is medical care, finance, education, or industrial manufacturing, urban transportation, etc., armed with high technology to accelerate the transformation of intelligence and shape new advantages in the new wave of intelligence, it has become the same choice for thousands of industries.
At the moment when the cloud economy explodes, as the infrastructure of the future digital economy era, the importance of the data center to the layout of artificial intelligence is self-evident.
Especially in the past two years, the speed of shifting of 5G technology, the technological transformation of the industrial Internet, the acceleration of "new infrastructure" policies, and the implementation of diversified digital applications are forming a powerful force that is driving the explosion of data centers. The development of nature also imposes higher requirements on the computing power infrastructure of the data center.
From the historical origin, artificial intelligence has experienced the budding from the early days of the 1950s, mainly using computers to prove mathematical problems, geometric problems and some mechanical intelligent expert systems; to the development of the 20th century, it is mainly used in some simple In terms of chess, machine translation and image recognition, it belongs to relatively primitive machine learning. It was not until the beginning of the 21st century that artificial intelligence ushered in a quantitative change under the catalysis of mature industries such as Moore’s Law, deep learning and big data applications. A qualitatively changed large-scale development by leaps and bounds.
At the time of the three ups and downs, the protracted journey of "hunting" and "gold rushing" artificial intelligence has also received explosive results in recent years. The process of integrating technology with mass consumer goods is the latest one.
At this point, artificial intelligence is no longer high and low.
The new battlefield of energy and environment AI revolution: data center
With the maturity of the algorithm model, the improvement of computing power, and the massive accumulation of data, AI technology has been widely used in mature and extensive scenarios in the fields of "new infrastructure", new retail, smart driving, smart cities, and smart manufacturing. Especially in the consumer Internet field, apps like Meituan, Douyin, and Ant Financial Services provide platform, content, and financial services that are actually supported by data assets that have been deposited over the years. Intelligent deep learning technology is trained to form core algorithms, which are then transformed into platform services.
However, in the field of energy and environment, because most of it was promoted by the government in the past, it was not as valued by capital as the consumer market, which made it still lagging in AI application scenarios. When competition in other fields became a red sea, capital looked back and saw that the application of AI in the energy environment was a blue ocean that had not yet been fully tapped, and the prospects were very broad.
In order to cater to the national strategy of carbon neutrality and realize the energy-saving emission reduction and carbon emission targets proposed by the state, it is urgent to carry out AI energy-saving in the field of energy and environment. In particular, the data center industry is becoming an important application area for AI energy saving.
As for how AI is applied in the data center field? How will it contribute to economic growth in the future? According to Professor Ma Xiaoming, the head of the School of Environment and Energy, Peking University, the current stage of artificial intelligence mainly uses deep learning algorithms to make predictions, combined with deep learning conclusions and expert experience, to give predictive maintenance recommendations and recommend the best Combination of equipment and early warning of fault judgment. According to Professor Ma Xiaoming, for the application of AI in the field of data centers, Peking University has launched in-depth cooperation with Vertiv Technology (Vertiv, NYSE: VRT) in production, education, and research to promote the implementation of AI technology in data center energy conservation as soon as possible.
Of course, at this stage, to achieve true artificial intelligence, data is undoubtedly the first element. As long as there is data, algorithm models can be used to mine and play useful value, making artificial intelligence a new driving force for social and economic development.
For example, Ant Financial can use big data to determine whether a loan can be granted to an individual. It can use massive databases for data mining, deposit high-value data assets, and then form its own set of risk control system.
Coincidentally, on October 31st, Google's DeepMind research team, which is Alpha Dog, also used artificial intelligence algorithms to predict the 3D structure of the DNA molecule spiral. But this research has very high requirements on the algorithm: it needs to consume 200 to 300 GPU devices to run training in a few weeks, and the cost of computing power is very high.
With the increasing scale of large data centers, the pressure on energy consumption has also increased, and the accumulation of massive amounts of data has also given birth to a set of solutions for data center energy saving using artificial intelligence technology. You know, in Google's big data center, there are millions of servers alone. Once running, the energy consumption generated in it can be imagined.
Modular data center thermal management AI optimization landing new results
In fact, as early as 2013, Google tried to use artificial intelligence to control water cooling to save energy in large data centers. They use neural network learning and training to accurately predict the PUE value and make energy-saving control solutions, which can ultimately achieve 15%-40% energy saving. Later, Tencent and Alibaba also optimized and followed up on Google’s plan, optimizing large-scale data centers, and achieved good results.
In this process, the advantage of artificial intelligence is that it can avoid some complex physical models in the middle through the advantages of algorithms and data, and treat the operation as a black box, and there is no need to care about the internal physical operation logic-in the end How to control the air conditioner? Just care about how much electricity is used for refrigeration? What is the input-to-output relationship between input environment variables?
Just use artificial intelligence technology to collect a large number of samples and learn this relationship. Once this input is given next time, the artificial intelligence can immediately determine how to control, in order to get the best output, so as to achieve automatic control.
For example, when the computing power demand of the data center is small, the server load pressure will be reduced, and the heat dissipation pressure will be relatively small at this time. After learning through the algorithm, some servers and air-conditioning and refrigeration equipment can be completely automatically turned off. In this process, due to the predictability of the algorithm, it can foresee changes and take actions in advance to make temperature changes smoother, thereby achieving optimization in energy-saving effects.
Driven by demand, in the process of evolution of the entire data center, many underlying logics will change. Some consider deployment speed and the ultimate full life cycle TCO, some care about the reliable operation of the business, some are concerned about the safety of the data center, some consider the return on investment, some require priority in computing power, and some pay attention to how fast and cost-effective .
In many cases, due to the differentiation of industry requirements, there are naturally a variety of differentiated pain points during the implementation of data centers, which has led to an increasingly strong demand for data center customization.
In addition, with the development of 5G and edge computing, data center AI energy saving has a distributed, fast and timely, low latency, and can match various micro data center computing power and distribution, and distributed multi-site collaborative management in 5G scenarios. And so on, the demand has become more and more intense. Therefore, the value of AI in data center energy saving is getting greater and greater.
Vertiv is an early manufacturer of comprehensive industry-university-research cooperation in the field of data center AI energy saving, and has achieved important research results.
It is worth mentioning that it is difficult for a customized data center to copy Google's plan, and corresponding models need to be established according to different conditions.
"In traditional public cloud data centers, super-large data centers like Tencent and Alibaba already have a variety of realistic and feasible solutions. After the logic training is completed, just leave them there and don’t care, because they just provide a public cloud data platform for others. It is not like the customized data center provided by Vertiv, which has to face different scenarios." In the interview, Xu Huali, a postdoctoral fellow at the Chinese University of Hong Kong, said frankly.
According to Xu Happy, the energy-saving solution based on AI technology developed by him in cooperation with Vertiv has entered the stage of acceptance. After a large-scale testing cycle, specific products will be implemented.
Unlike Google’s need to collect a large number of sample data in advance and powerful computing power for training, the advantage of Vertiv's use of AI to achieve data center energy saving is that it does not require long-term optimization like large-scale data center optimization. Time to collect a large amount of data for training and optimization, but through the reinforcement learning algorithm model, plus only a small number of samples, it can ensure that even if the customer deploys from scratch, it can be guaranteed to be based on the load operation in 3 to 6 days A series of data such as external climate and environmental conditions are learned, so as to realize the automatic control of the air-conditioning system. The optimal operating point of the system PUE can be found under the conditions of different load rates, and higher levels of energy saving than ordinary control algorithms can be realized.
From this perspective, once the application of AI in data center energy saving is mature in the future, manufacturers represented by Vertiv will bring an innovative change to their product development and production.
Judging from the current data, Vertiv Technology (Vertiv) ranks first in China's non-self-use data center market, precision air-conditioning market, and UPS market. This is also a key factor for Vertiv to seize the first position in the field of data center AI energy saving.
As for the future, as a benchmark enterprise in the field of AI energy-saving applications in data centers, what kind of boundary applications will Vertiv do in the field? How much influence does it create? What's the outlook? In fact, there is still a lot of imaginable space.
After more than half a year of research and development cooperation with Vertiv, Xu Huili believes that, on the one hand, Vertiv may be able to use the data center energy-saving project as an entry point to promote the establishment of realization through AI. The industry standard for data center energy saving, from data collection to model selection, provides some standardized means to measure and evaluate, thereby forming a stronger industry influence. On the other hand, it can also extend the internal control of the air-conditioning system from the air-conditioning switch control, combined with artificial intelligence technology to establish an enhanced model, so as to achieve better energy saving.
Of course, not just limited to energy saving, artificial intelligence can also achieve more applications in the data center, such as operation and maintenance, fault diagnosis, and early warning. The data itself can automatically detect data center failures, and generate reports after discovery and prompt what means to take The solution enables the intelligent operation of the data center in all directions.