Shulou(Shulou.com)06/02 Report--

In 2016, if you ask a person, what is artificial intelligence? He will probably answer-- I will, AlphaGo, the monster that defeats the human chess saint.

But in 2020, when you pull a person on the street and ask him what AI is, he is likely to start counting with you: smart speakers that children chat every day, AI cameras that remotely monitor the condition of the elderly, selfies of AI beauty in mobile phones, and car robots with voice calls out of cars. The algorithm that the information platform pushes all the time, the recommendation of the good things that are always guessed by e-commerce, the health code shown to the doorman, and the less congested commuter.

Have you noticed that AI seems like a giant elephant at this point? When it first entered human society, it was like an elephant walking into the bedroom. People either thought the elephant was fake or the bedroom was fake, as if something impossible had happened.

When it really becomes the base to support the operation of society, when it is everywhere, everyone seems to be like the people asked by the king in the Great Nirvana Sutra, and the people in the room think it is a big radish, a big fan, a big pillar and a small grass rope.

It may not be necessary and impossible for ordinary people to see the whole AI elephant, but the place we touch, the flowing blood, and the beating pulse may be the closest temperature of the intelligent age.

So what on earth is the blood that supports this huge object? I think most people will come to the same answer as I do, which is the computing power surging under the intelligent dome just as electricity was to the industrial age.

Of course, it has long been a clich é to mention the value of calculation today. But AI has changed from an illusory future technology to the existence of lifting thousands of things and being close to the texture of ordinary people's daily life, which naturally makes computing power itself undergo new forging and tempering.

When the way humans understand AI, release AI, and apply AI begins to change, computing is quietly changing.

Vientiane update: what kind of computing power is needed for new infrastructure

Some people will say that it is the kind of super-large data center that never goes down; others will say that it is the high-performance terminal processor that makes IoT smart and does not need frequent maintenance. All these touch the thick legs of the elephant "intelligent computing". And watching the whole picture of the Colossus, I'm afraid we have to start with the audience who live in symbiosis with the Colossus.

First of all, AI has changed from AlphaGo to a "giant elephant" invisible to everyone, and the objects in which it is used and applied have changed. From laboratories and Internet companies such as DeepMind, it has become a "traditional" heavy industry in city management, supermarket retail, manufacturing plants, finance and insurance, security education and so on. This means that the computing power that underpins AI must also change from satisfying academic model training to a productivity tool that helps thousands of industries to improve quality and efficiency.

This naturally brings new challenges to the computing industry, in addition to high concurrency, high throughput will inevitably lead to a substantial upgrade of computing performance, another thing that can not be ignored is the desire of heterogeneous data growth for heterogeneous computing.

We know that the combination of cloud and AI+5G has given birth to the explosive growth of the Internet of things, and the combination of software and hardware at the edge of the cloud is the fate of AI, while a variety of complex end-to-side hardware also requires the computing system to be able to handle data from various sources and structures. For example, if you want to direct traffic, you need to uniformly apply all kinds of data, such as road sensors, intelligent traffic lights, special cameras, and so on, in order to effectively cure the "traffic disease" of the city.

In addition, with the increase of AI application scenarios and tasks, as well as the expansion of terminal unstructured data, we must have a high degree of flexibility on the basis of the need for computing power to continue to upgrade. To put it simply, 95% of the data is unstructured data, including voice, images, videos, documents. If all of them are processed by general-purpose computing, it will undoubtedly bring a heavy load to the processor. It is tantamount to allowing all grades of a primary school to be mixed together, and it is estimated that the school administrators and teachers will be the first to collapse. Therefore, the computing architecture is naturally more scientific because it can effectively match different computing clusters in different task scenarios and for different data and business processes.

Although today, AI seems to have penetrated into many textures of life and become a general purpose technology (GPT), that is, it is widely deployed as a productivity resource to promote the overall rise of new industries and new organizations. In this context, the combination of AI and the industry has only just begun.

To take the most intuitive example, during the epidemic, telecommuting, telemedicine, universal waiting for live broadcast of Leishenshan on Mount Vulcan, intelligent Mini Program epidemic prevention, intelligent temperature measurement, and the emergence of various kinds of sprinkler robots were constantly refreshing the public's new understanding of AI temperature.

Xiangxiangdu River: AI Industry Penetration of Teng Computing

As we mentioned earlier, the penetration rate of AI and industry is only 4%. Under the wave of new infrastructure, AI is almost locked together with 5G, new energy, autopilot, cloud computing and other technical propositions.

In the complex industrial environment, how to infiltrate AI into every capillaries of the industry? I think the story of the three beasts crossing the river in the Yupa Sai Jie Sutra may represent the necessary capabilities in today's computing field-rabbits only float on the surface, horses may not reach the bottom of the river, and elephants must be able to set foot on the bottom of the river, thoroughly instilling the power of AI into the depths of the industry.

In 2019, Huawei launched the computing industry layout of "one cloud, two wings and two engines", in which Teng Teng became the core of the connection between the industry and AI. Today, at the Huawei analyst conference, a computational giant elephant named Teng Teng is beginning to become clear.

First of all, the super-performance computing cluster forged on the basis of Teng Teng processor constitutes the huge body of the "computing colossus".

Based on the Teng series AI processors, the Atlas computing cluster built by Huawei achieves scalable surging computing power support through the combination of boards, servers, clusters, etc. For example, Atlas 200, the size of half a credit card, can achieve 16 TOPS INT8 computing power, 200 pieces of object recognition in a single frame, what is the significance of arithmetic upgrade?

For example, during the COVID-19 epidemic, doctors were originally required to spend 10-15 minutes on image analysis to diagnose the disease, but with the help of AI recognition, it only takes 10 to 20 seconds, and the diagnosis accuracy is as high as 99%, greatly reducing the workload of medical staff.

It reflects the extremely high performance-to-price ratio, and also alleviates the tense situation of "too many people in short supply". This is the basic threshold for many ordinary industries to enter the era of intelligence.

Secondly, Huawei Atlas computing solution, and behind the efficient programming and execution framework CANN, one-stop AI development and management platform ModelArts, machine vision AI development kit HiLens, these constitute the limbs of the giant elephant that firmly supports the demand for intelligence in the industry.

If AI wants to land on the industrial side, it is obvious that it will not be able to package a bunch of processors by itself. Reliable technical solutions, easy-to-use technical tools, matching scene adaptation and so on, only when these are delivered to the industry, is it possible to complete an AI revolution in the industrial form.

So we can see that in the full stack and full scene AI of Teng, through the two major solutions of Atlas, the industrial hub, data center solutions can build AI infrastructure in cloud scenarios to meet the needs of high parallelism, high throughput and large-scale computing, and help hard core industries such as smart cities, smart medicine, and astronomical exploration. The intelligent edge solution faces end-to-side and edge-side, through cameras, drones, intelligent stations, etc., so that the public daily access to the campus, transportation, supermarkets, monitoring and so on, can touch AI in an easier way.

In order to enable various industries to apply the technology itself more efficiently and easily, Huawei chooses to leave complex problems to itself, making the development and deployment of AI easier through a variety of consistent development ports, as well as high-performance SDK, documentation and industry Demo, so that customers'AI business can be launched quickly. For example, with the help of MindStudio development platform, developers can quickly achieve software migration and performance tuning, saving a lot of time costs.

At this point, the integration of Yunbian AI into the limbs of the intelligent society, the continuous transformation of the energy supply industry, it has become possible.

Finally, with the help of the open source of the full scene AI computing framework MindSpore, Teng Teng handed over the keys of AI to the vast number of developers in the way of "teaching people to fish", catalyzing the big explosion of AI ecological innovation.

Huawei Teng is not limited to building itself into a "golden cake" of the smart computing industry, but also spares no effort to promote the social value of AI inclusive.

Not long ago, MindSpore, the full-stack and full-scene AI development framework, has implemented open source, delivering basic capabilities such as efficient development, powerful computing power and full-scene deployment in the Teng system to Qianhang Baiye.

As a result, enterprises and individual developers who were originally limited by problems such as high computing power, high threshold of difficulty, and scarcity of talents can be fully activated to release their own ingenuity in the Pengteng ecology, create AI applications with more industrial fine-grained and vertical availability, and further open up the global competitive advantage of Chinese industry AI.

Enabling ecology, of course, is a kind of water grinding time that "slow work can produce detailed work", but it also reflects the greater potential and possibility of the intelligent age. And only when the whole society participates in it, the value of AI will be continuously magnified and eventually transformed into the gift of the times to all. The meaning of Teng Teng's gift points to the distance.

Calculation, engineering, and ecology constitute the industrial penetration of the "AI Colossus". At this time, it may be possible to answer a question-Xiangxiang crossing the river, where is the other bank?

All things are at peace, only because there is an elephant.

To sum up briefly, the "AI Colossus" constructed by Teng Teng has created at least triple kinetic energy:

1. A lower threshold for development. Whether it is computing cost, development difficulty, personnel input, community resource acquisition, etc., Teng Teng's full stack and full scene AI has almost completed a capability output for the whole society and industry. If companies and developers eager to introduce AI and Nuggets AI are teaming up, Teng Teng has almost sent an integrated force that integrates tanks, output, assistance, medical care, and so on, fully armed from AI processors, development frameworks, development tools, open source platforms, and so on. It is a bit of an exaggeration to say that lying down and winning, but it definitely gives the industry a fuller and comfortable space to plan, deploy and create, and tell their own stories more easily in this intelligent upgrading war.

two。 Broader AI capabilities. Teng's full-scene system makes the comprehensive adaptation of AI scenarios in the cloud into an upgrade package for the industry. On the one hand, some blank areas can be nourished by AI, such as traditional factories, which are unable to be intelligent because of large-scale cost, computing power consumption and other concerns. With the help of Teng's edge computing scheme, it can promote the adaptation of AI in hardware terminals with a more benign cost model.

For example, in the cooperation with Southern Power Grid, route inspection is prone to accidents or occlusion, which requires real-time monitoring, real-time reporting and real-time analysis, so the requirement for computing power is also very high. While Atlas 200can be flexibly placed on some robots and cameras, the computing power can reach 16TOPS, but the average working power consumption is only 1 watt, eliminating the worry of energy efficiency, so it can be purchased in large quantities. Replace the manual inspection scheme to ensure the personal safety of industrial workers.

In addition, the cloud edge also expands the scope of adaptation of AI. With the unified architecture of Teng Teng, developers can apply cloud-trained models to intelligent terminals in various scenarios such as cameras, smart stations and edge servers through one-time development and multi-terminal deployment. A digital rich mine far beyond the mobile Internet era is waiting for gestation and competition.

3. Greener social well-being. If you want to make a living, you have to be a little green. This sarcastic remark, put in the computing industry, is a real well-being of the whole society. The digitization process has led to the construction of large-scale data centers, and according to surveys conducted by authoritative institutions, the total annual expenditure on power and cooling of servers in our country is more than 2 billion US dollars. Green environmental protection, PUE and renewable energy utilization will become one of the necessary indicators for AI applications. After all, no matter how smart it is, no one wants to live in a city with a bad environment.

Therefore, in the Teng computing ecology, while controlling the power consumption of power supply on the edge side to achieve millisecond wake-up and milliwatt dormancy, and through technical iteration in the data center to reduce the environmental pressure caused by digitization, the latest Atlas 900cluster has become the industry's first fully liquid-cooled AI cluster, with an area saving of 80% and electricity savings of 60%. PUE is less than 1.5%, which is lower than the industry average standard. While saving energy and reducing emissions, it also reduces the direct cost of operation and maintenance of the operator's computer room.

Not only the Jinshan and Silver Mountains of AI, but also the green waters and green mountains of all mankind. In the future, we will also see explosive growth in data, and ubiquitous computing also means exponential energy consumption. And from the technological orientation of Teng Teng to read the humanistic feelings, maybe the other side of wisdom is what we really want.

At this Huawei analyst conference, I was particularly moved by a sentence about Teng Teng-- the technology that has been forgotten by the end user is the best technology. In the history of science and technology, there are countless great achievements, such as alternating current, communications, the Internet, which we have long been accustomed to today, supporting the best of times at this moment. In the future, we will also witness that AI flows in the blood of modern life in a silent way. It is available to us at any time within easy reach.

By that time, people will no longer care what this elephant named AI looks like. They only remember that there was an elephant in Taiping days.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.