Myung Min is from Ao Fei Temple
Qubits reports | Official account QbitAI
Which city in China has more AI power?
Hangzhou surpassed Shenzhen to become second, and Beijing still sat firmly in the top spot;
Nanjing's first top five pushed Shanghai out of the top 4, ranking jointly in the first echelon (Top 5);
Jinan ranked among the top ten for the first time, and Chengdu returned to the list after 3 years;
Shenzhen, Suzhou, Guangzhou, hefei ranking changes have risen and fallen, but they are all dead set on defending the Top10 status.
The annual ranking of China's AI computing cities is here again.
As a new driving force for the development of the digital age, the size of computing power is part of the AI strength of a city.
Through the ranking, it can already be felt that the competition in THE AI computing power of major cities in China has become more and more fierce, and the situation of the domestic AI industry is thriving and a hundred schools of thought has initially taken shape.
At this juncture, where the tide of artificial intelligence will flow becomes crucial.
What is the development trend of AI computing power industry? How big is the market? How is the regional hash rate distributed?
More importantly, what will be the direction of AI development in the future?
At the just-concluded Artificial Intelligence Computing Conference (AICC), the answer may be available.
<h1 class="pgc-h-arrow-right" data-track="14" > the pace of domestic computing power infrastructure construction has accelerated</h1>
Compared with last year, this year's AI city hashrate ranking changes are still not small.
One of the most eye-catching cities is Nanjing.
In previous years, its highest ranking was only ninth, but now it can shake Shanghai's status as fourth for many years, and its strength cannot be underestimated.
You know, Beijing (Baidu, ByteDance, etc.), Hangzhou (Ali, NetEase, etc.) and Shenzhen (Huawei, Tencent, etc.) basically divided China's large technology giants and emerging AI unicorns.
With its own advantages in economic development, Shanghai has a very obvious agglomeration effect of artificial intelligence industry, with more than a thousand core artificial intelligence enterprises, and has maintained the first echelon of domestic AI cities for many years.
Bei, Hangzhou, Shenzhen, Shang, and Guangzhou also have unique advantages in attracting top talents and building an entrepreneurial ecosystem.
In such an environment, why can Nanjing catch up with 6 cities at a time?
The "2021-2022 Chinese Intelligent Computing Power Development Evaluation Report" released during the AICC Conference explains this to a certain extent:
There is policy support, talent reserves, and entrepreneurial ecological bonuses.
One of the most eye-catching is the Nanjing Intelligent Computing Center, which was officially completed and put into operation in July this year.
As the highest computing power intelligent computing center currently in operation in the Yangtze River Delta region, the computing power of Nanjing Intelligent Computing Center can reach 800P OpS, which is 8 billion billion operations per second.
It can process 10 billion images, 3 million hours of speech translation or 10,000 kilometers of autonomous driving AI data processing tasks in less than 1 hour.
This can undoubtedly inject new impetus into the development of Nanjing's artificial intelligence industry.
Moreover, it is oriented to many entities such as governments, enterprises and scientific research institutions, and plays a decisive role in accelerating the landing of science and technology finance, intelligent manufacturing, smart retail, smart medical care, smart transportation and other scenarios.
As a result, it is not surprising that Nanjing can directly enter the TOP5.
In fact, Nanjing is not the only one aiming at the ai development highland of the intelligent computing center.
This week, Wuxi announced a strategic signing with Inspur to build the Wuxi Intelligent Computing Center.
Hangzhou, Guangzhou, Dalian, Qingdao, Changsha, Taiyuan, Nanning and other places have also long put the construction of intelligent computing centers on the agenda.
Obviously, all regions in China are competing to lay out their own artificial intelligence computing centers, and the infrastructure construction of computing power has become the trend of the times.
<h1 class="pgc-h-arrow-right" data-track="34" > computing power should move towards multivariate integration</h1>
Through the appearance of major cities catching up with each other and healthy competition in the field of AI, it is not difficult for us to feel that the current AI industry is rising.
So, in addition to the trend of computing power infrastructure construction that has been seen, what characteristics will the development of the AI industry show?
Some references are also given in the "AI Computing Power Development Assessment Report".
The first thing to say is the diversification of AI scenarios.
The report divides many current scenarios by starting period, development period and mature period:
Among them, intelligent voice industries such as smart speakers and AI customer service have entered a mature period.
Intelligent manufacturing, smart medical care, smart retail and other fields are in the ascendant under the impetus of the epidemic.
Although the attention of autonomous driving is high, it has not yet fully entered the development period; smart scientific research has seen the hope of starting under the breakthrough of applications such as AlphaFold2 and China's Tianyan FAST.
Such a rich scene often requires a variety of computing types behind it.
As a result, the demand for chip diversification is becoming more and more obvious, and chip manufacturers are also trying to meet this demand as much as possible.
Today's dedicated-level chips on the market will not only consider the working purpose of the chip, the deployment environment, but also consider the application scenarios, so as to optimize and improve the running speed.
Like the Intel SG1 chip on display at the AICC conference, it is a processing card for cloud travel, video and images.
Domestic manufacturers such as Flinthara Technology and Horizon are also continuously launching special chips for application scenarios such as automatic driving and AIoT.
But then there are also some questions, that is, how to integrate multiple computing power?
While the application scenarios are moving towards diversification, they are also moving towards complexity.
In an AI application scenario, the amount of computation is often very large, and multiple tasks need to be completed, and a variety of chips are required to work together.
Therefore, how to make different chips cooperate with each other and release real computing power in the scene has become a major challenge.
In this regard, Liu Jun, vice president of Inspur Information, proposed at the AICC conference that the key to efficiently releasing the computing power of multiple chips is the platform.
From the manufacture of chips to large-scale use, there is often a huge gap in the industrial chain.
Taking the development of an AI server as an example, more than 280 key process control points and design problems need to be solved, and it is also necessary to achieve optimization and adaptation with algorithm frameworks and AI applications.
At the same time, the construction of large-scale AI computing power platforms is facing new problems of high power consumption, high current density, high bus rate, and high system complexity.
Therefore, creating an intensive, efficient, open and shared intelligent computing system is the key to making diversified computing power truly go to the industry and achieve universal and inclusive computing power.
<h1 class="pgc-h-arrow-right" data-track="57" creates greater value > ecological collaboration</h1>
From the above content, it is not difficult to see that the development of the domestic AI industry has shown a thriving trend.
And we can also really feel that AI is not far away from our lives.
This is often not a situation that can be formed by a single enterprise, but more rely on multi-party cooperation within the AI industry chain.
For example, the "meta-brain ecology" jointly formed by Inspur and its partners is a good example.
Through technology co-creation and resource sharing, Yuan Brain Ecology realizes the ability integration between partners, and can realize the delivery and service of the whole life cycle of intelligent transformation.
For example, based on the AI technology support of the meta-brain ecological platform, Ande Medical Intelligence has launched an AI-assisted diagnosis system for myocardial lesions, intracranial tumors, cerebral small blood vessel diseases and other diseases in artificial intelligence image analysis.
For example, Zhuoshi Zhitong uses the meta-brain ecological platform as a technical support to launch intelligent transportation AI systems such as traffic video analysis and traffic big data analysis platform.
In addition to the landing in the industrial ecology, this year we also saw that the computing infrastructure supplier made an attempt in algorithms and also embraced an open ecology.
In September, Inspur Artificial Intelligence Research Institute released the world's largest Chinese AI giant model "Source 1.0", with a parameter scale of 245.7 billion, trained by a 5,000 GB dataset, able to easily understand and create Chinese, can dialogue with humans, write poems according to propositions, write news, and continue stories.
At the AICC Conference, the "Source 1.0" open source program was officially released, targeting universities, meta-brain ecological partners and intelligent computing centers.
At the scene, Liu Jun, vice president of Inspur Information, introduced the original intention of the open source plan:
By opening up data, APIs, and code to the above 3 groups, Wave is hoping that in the open source environment, everyone can collide with more sparks.
At the same time, he also revealed that "Source 2.0" will also be available after that, which will be more targeted at multimodal, visual and other fields. And "Source 2.0" will pay more attention to cooperation, more joint exploration with the top companies in the industry.
<h1 class="pgc-h-arrow-right" data-track="71" what > do in the future? </h1>
Where will the tide of AI go?
From the above aspects, you can already feel one or two.
At the end of the report, IDC also made its own recommendations for different areas:
For technology suppliers: meet the needs of users to refine scenarios, build an open and compatible technology development path, start from the application needs of users, promote technological innovation, pay attention to green energy conservation, and empower the industry's double-carbon transformation.
Industry-oriented: Infrastructure construction and government guidance "two-pronged approach"; to create an open industrial ecology and promote the landing of diversified technological innovation.
For industry users: actively explore the application of artificial intelligence industry, optimize the input-output ratio, choose a technologically advanced, green and energy-saving computing infrastructure; and steadily promote AI computing capacity building to bring tangible value.
After watching this domestic AI "wind vane" conference, we feel that it has become the norm for major cities to catch up with ai, AI has gradually gone out of the laboratory, connected more enterprises, and formed its own unique ecology, and open source has become a unique magic weapon for development in the industry.
Where AI will go next is up to us to wait and see.
— Ends —
Qubit QbitAI · Headline signing
Follow us and be the first to know about cutting-edge technology developments