Computing power is productivity, and those who have computing power win the world.
With the rapid development and breakthroughs of the new generation of artificial intelligence technology, the demand for AI computing power, mainly based on deep learning computing models, has increased exponentially.
Data shows that between 1960 and 2010, the computational complexity of AI doubled every two years; between 2010 and 2020, the computational complexity of AI surged 10 times every year.
Faced with such large models and complex calculations, it is urgent to improve AI computing power.
After all, among the three elements of the development of artificial intelligence: data, algorithms and computing power, both data and algorithms are inseparable from the support of computing power.
So, where does AI computing power come from?
AI computing power has entered a period of "big construction"
The important role of AI computing power in improving the core competitiveness of national and regional economies has become an industry consensus.
The "2020 Global Computing Power Index Assessment Report" shows that for every increase in the computing power index by 1 point on average, the digital economy and GDP will grow by 3.3‰ and 1.8‰ respectively." The driving effect on the digital economy is visible to the naked eye. It has become the internal logic for national and local governments to increase their AI computing power.
Looking internationally, under the government investment-led model, the United States relies on the six national laboratories under the Department of Energy and the intelligence of universities supported by the National Science Foundation. The three major systems of the Computing Center and the Supercomputing Center, a research center affiliated to NASA, vigorously promote the computing platform to "grow" intelligent computing capabilities; the European Union has built 8 large-scale computing platforms in Europe through the "EuroHPC Plan" and actively promoted Combination with artificial intelligence technology.
my country’s AI computing power construction is also entering a period of “big construction”.
Since the State Council issued the “New Generation Artificial Intelligence Development Plan” in 2017, my country’s AI computing power construction is also entering a period of “big construction”. Through government investment-led, corporate investment-led, government-enterprise joint ventures and other models, a number of AI computing power infrastructures have been built. The construction of the center achieves effective resource structure integration at the national level. At present, eight places across the country have started to build national computing power hub nodes, and 10 national data center clusters have been planned to promote intensive, green, energy-saving, safe and stable computing. The construction of artificial intelligence computing centers.
At the same time, with the encouragement and support of new infrastructure and other policies, local governments are also actively carrying out the construction of artificial intelligence computing centers (AIDC).
According to intelligence According to East-West statistics, between January 2021 and February 2022, there are more than 20 artificial intelligence computing centers planned, under construction and put into operation across the country, of which 8 cities’ artificial intelligence computing centers have been completed and put into operation.
The computing power scale that AIDC in various places can provide or plan is generally 100PFLOPS, which is equivalent to the computing power of 50,000 high-performance computers.
For example, the first phase of Wuhan AIDC can provide 100PFLOPS of computing power. From the start of operation in May to December 2021, more than 100 companies have been attracted to settle in, more than 50 types of scenario-based solutions have been hatched, and the average daily computing power usage exceeds 90%.
In Wuhan AIDC With its computing power usage approaching saturation, it completed the second phase expansion project at the end of 2021, which can provide a total computing power of 200PFLOPS, and is planning a third phase expansion project.
In the field of AI industrialization, industrial AI and government Driven by demands such as intelligent governance, my country's AI computing power is booming.
According to the "2022-2023 China Artificial Intelligence Computing Power Development Assessment Report" jointly released by IDC and Inspur Information, China's AI computing power continues to Maintaining rapid growth, the scale of intelligent computing power will reach 268 exascale operations per second (EFLOPS) in 2022, exceeding the scale of general computing power.
It is expected that the annual compound growth rate of China’s intelligent computing power scale will reach 52.3%, while the compound growth rate of general computing power scale during the same period was 18.5%.
AI computing power construction still faces challenges
Currently, AI computing power construction is still in its early stages of exploration, with problems such as inconsistent construction standards and confusing industry pricing. These problems are becoming obstacles to the development of this new platform.
In the white paper "Research on the Development of New Generation Artificial Intelligence Computing Power Infrastructure", the National Industrial Information Security Development Research Center sorted out four issues in the construction of AI computing power infrastructure: National top-level system construction and standard system There is still no uniformity. Domestic AI chips and other software and hardware technologies are still subject to foreign restrictions. Large intelligent computing centers face duplication of construction and high energy consumption. The emphasis on construction and neglect of applications results in the inability to cover the needs of different application scenarios.
Taking the chaotic pricing standards in the industry as an example, Huang Peng, deputy chief engineer of the National Industrial Information Security Development Research Center and director of the Information Policy Institute, pointed out that the construction investment of two intelligent computing centers with similar functions and similar scales differed by more than 6 times.
From the perspective of the construction cost of AI computing power, it can be divided into infrastructure such as factories, equipment and products such as servers and computing chips, post-operation and maintenance costs, and electricity bills.
In terms of the cost of infrastructure, electricity bills, personnel, etc., there may be differences between the east and west, but the difference is not as much as 6 times.
This also shows that our country is still in the initial stage of the development of intelligent computing power and has not yet formed a standardized model.
Huang Peng believes that the construction of intelligent computing centers can refer to the computing power price standard plan launched by the Artificial Intelligence Industry-Academic-Research Innovation Alliance of the Chinese Academy of Sciences - in a series of comprehensive storage, energy consumption, development, customization, data scheduling, etc. After the factors are substituted into the clear algorithm standards, it is concluded that the basis of the intelligent computing center is to have 5P double-precision computing power (64-bit), 25P single-precision computing power (32-bit), and 100P half-precision computing power (16-bit). The facility price is approximately 100 million to 150 million yuan.
Huang Peng suggested that local governments should conduct sufficient research and demonstration before building an AI computing power platform to avoid "low-level duplication of construction" and "mismatch with local development needs."
If problems such as "conceptual confusion", "price confusion" and "emphasis on construction over application" can be avoided through sufficient research and demonstration, some deep-seated problems in AI computing power construction still require the entire industry chain. For example, AI software and hardware technology is restricted by foreign countries, the energy consumption of large-scale AI computing centers is too high, and the cost is too high, etc.
The enterprise level must actively explore AI technology research and development and model innovation, especially increase independent research and development of core technologies such as AI chips, work with upstream and downstream to improve industrial chain cohesion and ecological compatibility, and at the same time strengthen The research and development of software platforms and applications such as algorithms, frameworks, and models puts technology development in your own hands.
Scenario implementation promotes the development of AI computing power
It is undeniable that the larger the computing power of a country, the higher the level of economic development.
There has been a significant positive correlation between the computing power scale and the level of economic development in various countries around the world.
The arrival of the era of intelligent interconnection of all things and the implementation of AI intelligent scenarios will generate unimaginable massive amounts of data. These data will further stimulate demand for AI computing power.
The "2022-2023 China Artificial Intelligence Computing Power Development Assessment Report" shows that the top five industries with application penetration in China's artificial intelligence industry in 2022 are the Internet, finance, government, telecommunications and manufacturing.
Compared with 2021, the penetration of AI in the industry has increased significantly.
Among them, the Internet industry is still the industry with the highest penetration and investment of artificial intelligence applications;
The penetration of artificial intelligence in the financial industry has increased from 55% in 2021 to 62%, and intelligent customer service, Physical robots, smart outlets, cloud access points, etc. have become typical applications of artificial intelligence in the financial industry;
The penetration of artificial intelligence in the telecommunications industry will increase from 45% to 51% in 2021, and artificial intelligence technology will be integrated into telecommunications networks construction, optimization, and provide support for the construction of the next generation of smart networks;
The penetration of artificial intelligence in the manufacturing industry has increased from 40% to 45%. It is expected that by the end of 2023, 50% of China’s manufacturing supply chains Artificial intelligence will be used in this link.
As the usage and development of new technologies and application scenarios continue to increase, it has also brought continuous impetus to the development of AI computing power.
First, cloud AI models are developing in a large-scale direction, and the construction of computing power infrastructure has become a key element of competition.
The BERT large model launched by Google in 2019 has 340 million parameters, uses 64 TPUs, and costs $15,000 to train to target accuracy.
In 2020, the GPT-3 large model launched by OpenAI has 175 billion parameters and the training cost reached US$12 million.
In 2021, Microsoft and NVIDIA used 4,480 GPUs to train a large MT-NLG model with 530 billion parameters, and its training cost was as high as US$85 million.
Second, the demand for AI computing power at the edge is increasing rapidly.
Emerging application scenarios such as cloud gaming and autonomous driving have put forward higher requirements for the speed and magnitude of data transmission. Edge terminals are deployed between terminals and clouds to form a "cloud-edge-end" communication architecture. Become the main direction of future technological development.
The growing demand for edge computing will effectively drive the development of AI computing power.
Third, intelligent connected cars have increased demand for AI computing power.
The penetration rate of self-driving cars continues to increase, and car driving control systems are developing towards intelligent functions such as "perception-recognition-interaction".
Therefore, the intelligent driving AI model plays an important role in helping the car move from the L1/L2 assisted driving stage to the L3/L4 autonomous driving stage, and ultimately realize the functions of "intelligent decision-making and real-time control".
With the increase in the amount of various types of driving data that need to be trained, and the increasing demand for developing intelligent driving AI models, the demand for AI computing power will increase significantly in the future.
Fourthly, the construction of the virtual reality world requires the support of AI computing power.
AI technology provides intelligent empowerment such as modeling automation and intelligent interaction methods for the virtual reality world, which is expected to improve the efficiency of VR content production and the immersive experience of users.
VR content providers’ pursuit of quickly creating virtual scenes and improving user sensory experience has stimulated their demand for AI computing power.
Conclusion
A new round of computing power revolution is accelerating.
As a new productive force, computing power still faces many challenges. It is necessary to implement the inclusive function of centralized computing power and truly play the role of "electricity" and "oil".
After all, computing resources that are “available, affordable, and used well” are the real infrastructure that changes productivity.
The above is the detailed content of In the era of 'big construction”, where will AI computing power go?. For more information, please follow other related articles on the PHP Chinese website!

机器学习是一个不断发展的学科,一直在创造新的想法和技术。本文罗列了2023年机器学习的十大概念和技术。 本文罗列了2023年机器学习的十大概念和技术。2023年机器学习的十大概念和技术是一个教计算机从数据中学习的过程,无需明确的编程。机器学习是一个不断发展的学科,一直在创造新的想法和技术。为了保持领先,数据科学家应该关注其中一些网站,以跟上最新的发展。这将有助于了解机器学习中的技术如何在实践中使用,并为自己的业务或工作领域中的可能应用提供想法。2023年机器学习的十大概念和技术:1. 深度神经网

实现自我完善的过程是“机器学习”。机器学习是人工智能核心,是使计算机具有智能的根本途径;它使计算机能模拟人的学习行为,自动地通过学习来获取知识和技能,不断改善性能,实现自我完善。机器学习主要研究三方面问题:1、学习机理,人类获取知识、技能和抽象概念的天赋能力;2、学习方法,对生物学习机理进行简化的基础上,用计算的方法进行再现;3、学习系统,能够在一定程度上实现机器学习的系统。

本文将详细介绍用来提高机器学习效果的最常见的超参数优化方法。 译者 | 朱先忠审校 | 孙淑娟简介通常,在尝试改进机器学习模型时,人们首先想到的解决方案是添加更多的训练数据。额外的数据通常是有帮助(在某些情况下除外)的,但生成高质量的数据可能非常昂贵。通过使用现有数据获得最佳模型性能,超参数优化可以节省我们的时间和资源。顾名思义,超参数优化是为机器学习模型确定最佳超参数组合以满足优化函数(即,给定研究中的数据集,最大化模型的性能)的过程。换句话说,每个模型都会提供多个有关选项的调整“按钮

截至3月20日的数据显示,自微软2月7日推出其人工智能版本以来,必应搜索引擎的页面访问量增加了15.8%,而Alphabet旗下的谷歌搜索引擎则下降了近1%。 3月23日消息,外媒报道称,分析公司Similarweb的数据显示,在整合了OpenAI的技术后,微软旗下的必应在页面访问量方面实现了更多的增长。截至3月20日的数据显示,自微软2月7日推出其人工智能版本以来,必应搜索引擎的页面访问量增加了15.8%,而Alphabet旗下的谷歌搜索引擎则下降了近1%。这些数据是微软在与谷歌争夺生

荣耀的人工智能助手叫“YOYO”,也即悠悠;YOYO除了能够实现语音操控等基本功能之外,还拥有智慧视觉、智慧识屏、情景智能、智慧搜索等功能,可以在系统设置页面中的智慧助手里进行相关的设置。

人工智能在教育领域的应用主要有个性化学习、虚拟导师、教育机器人和场景式教育。人工智能在教育领域的应用目前还处于早期探索阶段,但是潜力却是巨大的。

阅读论文可以说是我们的日常工作之一,论文的数量太多,我们如何快速阅读归纳呢?自从ChatGPT出现以后,有很多阅读论文的服务可以使用。其实使用ChatGPT API非常简单,我们只用30行python代码就可以在本地搭建一个自己的应用。 阅读论文可以说是我们的日常工作之一,论文的数量太多,我们如何快速阅读归纳呢?自从ChatGPT出现以后,有很多阅读论文的服务可以使用。其实使用ChatGPT API非常简单,我们只用30行python代码就可以在本地搭建一个自己的应用。使用 Python 和 C

人工智能在生活中的应用有:1、虚拟个人助理,使用者可通过声控、文字输入的方式,来完成一些日常生活的小事;2、语音评测,利用云计算技术,将自动口语评测服务放在云端,并开放API接口供客户远程使用;3、无人汽车,主要依靠车内的以计算机系统为主的智能驾驶仪来实现无人驾驶的目标;4、天气预测,通过手机GPRS系统,定位到用户所处的位置,在利用算法,对覆盖全国的雷达图进行数据分析并预测。


Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

AI Hentai Generator
Generate AI Hentai for free.

Hot Article

Hot Tools

Dreamweaver Mac version
Visual web development tools

MinGW - Minimalist GNU for Windows
This project is in the process of being migrated to osdn.net/projects/mingw, you can continue to follow us there. MinGW: A native Windows port of the GNU Compiler Collection (GCC), freely distributable import libraries and header files for building native Windows applications; includes extensions to the MSVC runtime to support C99 functionality. All MinGW software can run on 64-bit Windows platforms.

MantisBT
Mantis is an easy-to-deploy web-based defect tracking tool designed to aid in product defect tracking. It requires PHP, MySQL and a web server. Check out our demo and hosting services.

Atom editor mac version download
The most popular open source editor

Notepad++7.3.1
Easy-to-use and free code editor
