Feature article｜The demand for computing power explodes under the boom of AI large models: Lingang wants to build a tens of billions industry, and SenseTime will be the "chain master"

SenseTime launched the operation of the Artificial Intelligence Computing Center AIDC in the Lingang New Area of ​​the Shanghai Free Trade Zone on January 24, 2022. At that time, the artificial intelligence company may not be able to accurately predict that 2022 will become the first year of the so-called AIGC (AI Generated Content).

“Today, our Lingang AIDC has nearly 30,000 GPUs (graphics processing units), and our current computing power has reached 5,000 PetaFLOPS (1 PetaFLOPS is equal to 1 quadrillion floating point operations per second). On top of this, we believe that we can have better developer efficiency in the future and be able to support more large-scale model computing power training with a scale of hundreds of billions." On June 2, "AI leads the era, computing power drives the future" - — Xu Li, chairman and CEO of SenseTime Technology, said at the Lingang New Area Intelligent Computing Conference.

SenseTime told reporters from The Paper (www.thepaper.cn) that there are still many demands waiting to be processed. According to Yang Fan, co-founder of SenseTime and president of the large device business group, artificial intelligence’s pursuit of larger data, larger scale, and greater computing power did not “start today.” “The entire artificial intelligence technology iteration, The history of progress can be seen as a pursuit of 'aesthetics of violence' and a technological iteration process in which the three elements of algorithm, computing power and data change from quantitative to qualitative."

The Lingang New Area, which focuses on cutting-edge industries, responded quickly to this new craze. On June 2, Wu Xiaohua, deputy secretary of the Party Working Committee of Lingang New Area, released the "Action Plan for Accelerating the Construction of Computing Industry Ecosystem in Lingang New Area" at the above-mentioned conference. Under the blueprint of the "Plan", by 2025, Lingang will become a computing power industry cluster with national influence, and the overall scale of the computing power industry, including related hardware, software, applications, services, etc., will exceed 10 billion yuan.

Wu Xiaohua, deputy secretary of the Party Working Committee of Lingang New Area, released the "Action Plan for Accelerating the Construction of Computing Industry Ecosystem in Lingang New Area".

"We see that the era of AI explosion has arrived. AI has entered all areas of our production and life. With the explosion of AI applications, it has actually driven an explosion in the demand for computing power." For After the introduction of the above-mentioned "Plan", Lu Yu, director of the High-tech Division of the Lingang New Area Management Committee, told media including The Paper (www.thepaper.cn) that Lingang already had good advantages in the early stage, "That's us. The computing power resources are very abundant.”

More importantly, when artificial intelligence companies choose whether to land in Lingang, computing power resources have become a particularly important decision-making factor.

Computing power is the energy of the new era, and success lies not only in “aesthetics of violence”

What is computing power? Xu Li believes that computing power is actually an expression of the entire model’s capabilities. “Computing power is equal to the parameters of the algorithm or large model, multiplied by the amount of data it processes. In the era of large models, the larger the parameters, the greater the amount of data multiplied. , the greater the computing power required." Computing power has become the energy source of the new era. "To a certain extent, computing power determines the competitiveness of the market."

Yang Fan also mentioned that from last year to this year, a very popular concept in the field of artificial intelligence is called content generation. At the same time, everyone is familiar with the term large model. To put it simply, this is a "violent aesthetic". For example, the GPT-3 model uses more than 175 billion parameters and requires a high-performance processor to support training. The V100 training consumes 10,000 cards for 14.8 days. , the overall computing power requirement is about 625 PetaFLOPS.

Yang Fan believes that this “aesthetics of violence” can also be understood as quantitative changes leading to qualitative changes. “In fact, from the first day of its birth to today, artificial intelligence has been pursuing greater intelligence through scale. " He mentioned that in fact, in the field of artificial intelligence, in the past 5-6 years, the consumption of computing power by the industry's top artificial intelligence models "has doubled every 4-6 months. , which means that it has increased nearly 300,000 times in the past few years.”

Of course, "violence" and "intelligence" are not completely proportional. "Having larger resources and a larger scale is only a necessary but not sufficient condition." Yang Fan emphasized that the real "aesthetics of violence" "What supports major technological innovations and achievements lies in the continuous optimization and improvement of every link.

Take data as an example, "The data used by GPT-4 is actually only 1% of all the data collected by OpenAI, because he found that when more data is fed into the robot, it may become less smart. More effective and higher-value data should be provided to this algorithm, and then a smarter brain can be created."

It believes that, at least today, the validity of data is far more important than the total amount of data. As for how to define effective data, "This actually requires a lot of efforts from data scientists. OpenAI actually lets their best scientists do the data, not the algorithms as everyone thinks."

This kind of optimization of every link also includes computing power. When Nvidia is out of stock, why is no one using domestic chips for commercial large-scale training? Why did Nvidia make all the money immediately after the latest wave arrived? The explanation behind these questions is, "It is not that we can generate the final value by stacking the computing power to a certain value. Putting 1,000 cards and 100 servers together to run the same task requires a lot of supporting software and communications. Network, it is a series of joint optimization processes of software and hardware. We have not done this kind of work in the past, and we need to make up for it today."

Following the trend, Lingang accelerates the formation of a diversified computing power supply system

According to Wu Xiaohua, the computing power industry in Lingang New Area has made corresponding arrangements in upstream software and hardware, midstream data centers, dispatching platforms, and downstream applications. Currently, the total computing power of Lingang exceeds 3EFLOPS (FP32, 1EFLOPS is equal to one second) 10 billion floating-point operations), of which intelligent computing power accounts for nearly 80%, and the total computing power accounts for nearly 20% of Shanghai.

The aforementioned "Plan" proposes that by 2025, the new area will form a multiple computing power supply system that focuses on intelligent computing power and coordinates basic computing power and super computing power, with a total computing power exceeding 5EFLOPS (FP32). The proportion of AI computing power has reached 80%, and the overall scale of the computing power industry (including related hardware, software, applications, services, etc.) has exceeded 10 billion yuan. A public computing power service platform has been established, the computing power trading mechanism has been standardized, and regional computing power scheduling has been realized. Create a computing power industry cluster with national influence and build a number of benchmark scenarios for computing power demonstration applications.

“Intelligent computing power is what the most popular AI companies need at the moment. We have also found that when AI companies come to Lingang, they no longer just focus on how much policy support and subsidies they can give them. Pay attention to whether the implementation here can solve his computing power needs, because computing power is now in very short supply in the market." Lu Yu mentioned this significant change.

SenseTime said that as of May this year, SenseTime large devices have provided services to more than 40 core customers. "Especially under the wave of large models, we now support more than 10 institutions to train their large models in the intelligent computing center in Lingang." Yang Fan also mentioned.

Shenzhen Technology, founded in 2018, is one of the demanders for computing power. The company's core team is led by E Weinan, an academician of the Chinese Academy of Sciences, and others. It is a pioneer in the "AI Science" scientific research paradigm. Its first "multi-scale modeling machine learning high-performance computing" paradigm has achieved a breakthrough in multi-scale molecular modeling. Unifying accuracy and efficiency in simulation.

According to previous reports by reporters from The Paper (www.thepaper.cn), Shenzhen Technology has launched the Lebesgue scientific computing platform, the Hermite drug design platform, and the Bohrium microscopic computing and design platform. For example, in the field of medicine, Shenzhen Technology has joined hands with many customers to integrate the computing paradigm of physical modeling AI with preclinical drug research and development more broadly. Through Hermite Uni-FEP, Uni-Fold, RiD and other modules, free energy microscopy Combining interference theory, molecular dynamics, enhanced sampling algorithms and high-performance computing to accurately predict protein structure and conformational changes, and efficiently evaluate the binding free energy of proteins and ligands with chemical precision, providing efficient and accurate theoretical guidance for drug developers , improve the efficiency of drug design and optimization.

On December 29 last year, Beijing-based Shenzhen Technology registered and established Shenzhen Energy Biotechnology (Shanghai) Co., Ltd. in Lingang. In an interview with a reporter from The Paper (www.thepaper.cn), Liu Huishi, Vice President of Government and Enterprise Affairs of Shenzhen Technology, said that the company’s deployment of a new generation molecular simulation algorithm research and development center and an AI-assisted drug design business center in Lingang was mainly due to the Da Lingang is vigorously developing computing power. "We have a demand for computing power during the training of the model. In addition, Lingang is especially vigorously developing localized computing power. We also want to contribute to this aspect."

We mainly conduct drug research and development business layout in Lingang, including the research and development of our own drug pipeline. " Liu Huishi mentioned that Shenzhen Technology's business has a positive and direct cooperative relationship with leading industries such as artificial intelligence and biomedicine in Lingang and even Shanghai. "We are willing to incorporate our R&D and products into the large-scale Lingang Come into the ecosystem. ”

The above-mentioned "Plan" also mentioned that the Lingang New Area has also formulated a series of safeguard measures, including strengthening talent protection, improving support policies, and promoting open cooperation. According to Lu Yu, if AI companies come to Lingang, they will give priority to Lingang’s intelligent computing power, and at the same time, through the issuance of computing power coupons and other forms, companies will be able to use computing power at a preferential price. “Even for key AI companies, the government will The computing power cost can be directly subsidized by no more than 30%, and we will come up with these policies.”

At this conference, China Telecom’s Lingang public intelligent computing service platform and domestic GPU joint innovation base were also officially launched, which is worth noting. China Telecom established Lingang Computing Power (Shanghai) Technology Co., Ltd., which will carry out the construction of Lingang Computing Power Park and launch 40,000 high-power racks suitable for intelligent computing and supercomputing in batches.

Tang Wenkan, deputy director of the Shanghai Municipal Economic and Information Commission, said on the same day that currently, the new generation of information infrastructure with "network as the basis, data as the core, computing power as the key, and security as the bottom line" has become an important step in building modern industries. Basic support. Shanghai has proposed to build a "2 (3 6) (4 5)" modern industrial structure, which places higher demands on the construction of new information infrastructure represented by computing power.

On May 16, the Shanghai Municipal Commission of Economy and Information Technology announced the list of data center projects that have passed the compliance assessment of the "Shanghai Data Center Construction Guidelines", supporting a total of 16 projects, including one located in Lingang There are 2 projects. "So far, our committee has supported 8 projects in the new area, including SenseTime AIDC, Youfu Networks, and Information Feiyu, with a total of 28,000 6kW standard cabinets, accounting for nearly 1/5 of the city's approved cabinets."

Tang Wenkan suggested making full use of Lingang’s computing resources and establishing public computing services. "At present, Lingang's SenseTime AIDC has been connected to the public computing service platform. I also hope that all units participating in the meeting today, especially telecom operators, will actively build extremely fast computing in Lingang based on the network characteristics of Lingang. The power carrying network helps realize the ubiquitous network, ubiquitous computing power, and omnipresent intelligence, and promotes computing power to become a public service like water and electricity."

Established the Intelligent Computing Industry Alliance, with SenseTime becoming the leader of the industry chain

Lingang’s goal is to create a computing power industry alliance that integrates upstream, midstream and downstream, which can meet current and future needs in a collaborative and systematic manner and utilize existing advantages to achieve development.

Lu Yu regards Lingang’s computing power supply as the “middle section” of the entire industry chain. One end provides computing power guarantees for AI companies that have settled in Lingang, and the other end involves computing power that is extremely critical. "Chips, software, systems", "We hope that there will be a demand side and such a platform side, then we will gather computing chip companies, software companies, and system companies here to let them deeply participate in such a system. During the construction process."

Yang Fan also emphasized, “The development of all the achievements of large models seen today is not only the miracle of violence, the improvement of technical value brought about by the continuous increase in the scale of the three elements of artificial intelligence, but also the basic research and development capabilities and The in-depth integration of the system's engineering capabilities, algorithm optimization, data sorting and selection, and optimization of platform computing power. These three are often interconnected, and it is difficult to turn them into separate links and do it alone. ”

He said that the important value of the intelligent computing power industry chain is that “Only when there are more companies in the chain, and everyone promotes mutual exchanges and thinking, and carries out more in-depth cooperation, can we be able to compete in such a new critical situation. To achieve better technological progress and support in the major technological wave."

At the conference site, the New Area Intelligent Computing Industry Alliance was formally established, and China Unicom was appointed as the alliance’s first rotating chairman unit. It is reported that China Unicom will establish the Yangtze River Delta Innovation Research Institute in the new area in the future to further assist the development of the intelligent computing industry in the new area.

Members of the Intelligent Computing Industry Alliance in the New Area are represented by companies providing computing power such as intelligent computing power, basic computing power and supercomputing power centers, GPU, FPGA, ASIC and other computing power chip companies, as well as large model, AI for science It consists of a total of 25 companies that require computing power, as well as three universities and research institutes, including the East China Branch of the China Academy of Information and Communications Technology, Xi'an University of Electronic Science and Technology of China, and University of Electronic Science and Technology of China. Resource sharing, technical exchanges, and project cooperation will be carried out in the future. SenseTime was awarded the title of “Intelligent Computing Industry Chain Leader in the New Area”.

GPU chip manufacturer Mu Xi said on the same day that three types of GPU products that meet AI inference computing, AI training/general computing, and high-performance rendering functions can be used in AI inference, AI training, data centers, metaverse, and cloud games. and other fields, will empower the transformation and development of various fields.

Tang Wenkan also has high hopes for the establishment of the Intelligent Computing Industry Alliance in the Lingang New Area. “Relying on chain owners such as SenseTime, combined with its own advantages, we will explore the collaboration of all elements of the upstream and downstream of the industrial chain to form a new digital economy. Breaking point."

12 companies jointly signed a collaborative procurement agreement for upstream and downstream enterprises in the intelligent computing industry in the new area at the conference that day. Lu Yu mentioned that the new area will also issue a positive list of collaborative procurement. “If companies purchase upstream products such as domestic GPUs during the process of building a localized computing platform, we will provide subsidies. This will also encourage upstream and downstream companies to upgrade. Good cooperation."

The above is the detailed content of Feature article｜The demand for computing power explodes under the boom of AI large models: Lingang wants to build a tens of billions industry, and SenseTime will be the "chain master". For more information, please follow other related articles on the PHP Chinese website!