The hottest Huawei released its overall computing

  • Detail

Huawei first released its overall computing strategy and launched the world's fastest AI training cluster atlas 900

on September 18, the fourth Huawei full connectivity Conference opened in Shanghai. Huawei first released its computing strategy, which is based on architecture innovation, investment in the full scene processor family, business strategies to do something, and building an open ecosystem to layout, thus opening a new era of great navigation in the computing industry. At the same time, Huawei launched atlas 900, the world's fastest AI training cluster, to accelerate the intelligent process of scientific research and business innovation

Hu houkun, vice chairman of Huawei, delivered a keynote speech

the main points of Hu houkun's speech are as follows

two years ago, Huawei released a new vision and mission, which is to bring the digital world into everyone, every family, every organization, and build an intelligent world with interconnected things

we believe that the intelligent world has three characteristics, namely, the perception of all things, the interconnection of all things, and the intelligence of all things. To support such an intelligent world, there are two key technologies that need continuous innovation and Investment: connection and computing

two key technologies in the intelligent world

Huawei has always been doing connectivity. In the past 30 years, Huawei has continuously invested in the field of connectivity, from wired to wireless, from 2G, 3G, 4G, to today's popular 5g, and Huawei has made some achievements

however, Huawei is not just a connection company. In our view, connectivity and computing are indispensable to support the future intelligent world. In fact, the two technologies of connection and computing, like a pair of twin brothers, promote and coordinate each other

it can be said that in the future intelligent world, where there is connection, there will be calculation and connection. In the past, we talked a lot about connection. In fact, Huawei has invested in computing for ten years. Today, I would like to share Huawei's insight and strategy in computing

computing has become an extension of human capabilities

first of all, let's talk about our observations on the computing industry. Since the first computer was born in 1946, for more than 70 years, from mainframe to personal computer, from desktop to notebook, to tablet, from intelligent to wearable devices, computing devices have become smaller and smaller, more powerful, and closer to people

it can be said that computing power has become an extension of human ability. Following this trend, we believe that the development prospect of the computing industry is unlimited

the computing mode has been evolving

over the past 70 years, the computing mode itself has also been developing. I first came into contact with computers from college. At that time, I studied programming. What the teacher taught was that if you give a formula, it will give you a result. In fact, this is a rule-based calculation mode. Such a calculation mode solved the problems of census and trajectory calculation well at the beginning

however, for some problems that cannot be described by certain rules, such as speech recognition, image recognition, real-time translation, etc., it is difficult for rule-based computing mode to solve these problems

fortunately, scientists have found a new calculation mode, that is, the calculation mode based on statistics, which can solve these problems mentioned above by regularly checking and revising the experimental machine. Moreover, this statistical based computing model has also become the key cornerstone of the current development of artificial intelligence

we believe that this statistical based computing model will gradually become the mainstream. It is estimated that in five years, the computing power consumed by AI computing will account for more than 80% of the total computing power consumption of the whole society. Based on such changes, we believe that computing is entering a new era of intelligence

computing has entered the intelligent era

in the intelligent era of computing, we believe that there are three important characteristics

the first feature requires super computing power. Statistical computation itself is a kind of violent computation, which is highly dependent on computational power. For example, in order to let the computer know a cat, it needs millions of pictures to train Abbott, which consumes amazing computing power. For more complex scenes such as autonomous driving, astronomical exploration, weather prediction, etc., the demand for computing power will be greater

the second feature is that computing and intelligence will be everywhere, not just on the central side. From violent computing at the central node to Professional Computing at the edge, such as gene sequencing, and personalized computing at the end, such as headphones and, together, constitute the computing form of the future intelligent era

the third feature is the need for efficient collaboration between end-to-end clouds. The center side is responsible for the calculation of the general model, providing collaborative support for the personalized calculation of the end side and the specialized calculation of the edge side

2trillion computing industry blue ocean

to achieve these three characteristics, we still face many challenges. For example, super computing power, exploring new computing architectures, full scene processors, and so on

in Huawei's view, challenges just mean opportunities. The greater the challenge, the greater the opportunity. What we are facing is a big blue ocean of computing industry. Here, Gartner's data is quoted. It is estimated that the scale of the computing industry will exceed $2trillion by 2023

Huawei's computing industry strategy

therefore, Huawei firmly invests in the computing industry. We mainly plan from four aspects, including breakthroughs in architecture innovation, investment in the full scene processor family, adhering to the business strategy of doing something and not doing something, and sparing no effort to build an open ecosystem

architecture innovation

from our previous insight, computing is everywhere and intelligent in the future, of which computing power is the key foundation

at present, the supply of computing power in the industry is scarce. The key to the supply of computing power lies in the efficiency of processors. At present, when Moore's law has almost reached its limit, from the perspective of industrial development, we must seek a breakthrough in the processor architecture and use a new processor architecture to match the growth rate of computing power

on the other hand, from the perspective of Huawei's own business layout, there are both network services and terminal services, as well as public cloud services, which comprehensively cover the end, edge and cloud. Therefore, full scene intelligence itself is the basic requirement of Huawei's business intelligence

therefore, we launched the Da Vinci architecture to realize the ubiquity of computing and intelligence, which is a natural choice. So far, Da Vinci computing architecture is the only processor architecture in the industry that can cover the whole scene of end, edge and cloud, which is also a solid foundation for Huawei to build the computing industry

investment in full scene processors

processors are the most basic part of the entire computing industry. After years of investment efforts, Huawei has released several series of processors

specifically, it includes Kunpeng series supporting general computing, shengteng series supporting AI, Kirin series supporting intelligent terminals, and Honghu series supporting smart screens. In the future, we will continue to invest in processors, and we will launch a series of processors for more scenarios

business strategies do something and don't do something

our business strategy is summarized in one sentence. In general, we don't sell processors directly to the outside world. We focus on cloud services for customers, components for partners, and give priority to supporting partners in developing complete machines. Specifically, we have three considerations:

Hardware opening: we open the server motherboard, AI modules, boards and other hardware to our partners to help our partners do a good job in the whole machine and solutions

open source software: we open source server operating system, database, AI development framework and other software to help partners make commercial versions and make software development easier

enable application development and migration: we don't do applications, but we provide tools and teams to help partners do application development and migration more efficiently

we hope to better support our partners through such a business strategy of doing something and not doing something

build an open ecosystem

finally, the computing industry is an open industry, which is highly dependent on ecology and requires global collaboration

in 2015, Huawei first released the fertile soil plan. In the past four years, we have made great progress, and have developed more than 1.3 million developers and more than 14000 ISVs

today, we will upgrade the fertile soil plan, continue to invest 1.5 billion US dollars, expand the scale of developers to 5million people, and enable global partners to develop applications and solutions. On the third day of this conference, my colleagues will release a detailed plan

general computing business strategy

in the field of general computing, we invest in Kunpeng series to build competitive general computing processors

on this basis, we continue to invest in key technologies and products such as boards, servers, operating systems, databases, compilers, etc. we hope that through strong investment, we can open up the whole ecological chain and let our partners participate in building Kunpeng industry with more confidence and firmness

we are also working with our partners to build Kunpeng industrial ecological base

together with local governments and partners, we will combine local advantages to build an innovation base and Incubation Platform for Kunpeng computing industry. Take the platform as the carrier, gather ecological partners, carry out application demonstration, cultivate industrial talents, and incubate industrial standards

at present, the work is carried out very smoothly. We have begun to implement the implementation in Beijing, Shanghai, Shenzhen and other cities. It is expected that more partners will join us in the future

ai Computing: full stack and full scene AI solutions were fully implemented

at the full connection conference last year, Huawei released a full stack and full scene AI solution. Frankly speaking, at that time, we only delivered the shengteng 310 processor and modelarts application development platform for reasoning

this year, our Pentium processor for training and mindspire, an AI computing framework, have been released. So far, Huawei's full stack and full scene AI solution has been fully implemented, and you can rest assured to cooperate with us

launch atlas 900: the world's fastest AI training cluster

today, we will launch a heavyweight product atlas 900, which integrates Huawei's decades of technology precipitation. It is currently the world's fastest AI training cluster, consisting of thousands of Pentium processors

in the resnet-50 model training, the gold standard for measuring AI computing power, Atlas 900 completed the training in only 59.8 seconds, which is 10 seconds faster than the original world record. What is this concept? It is equivalent to the sprint champion finishing the finish line and drinking a bottle of water before waiting for the second place

we believe that the powerful computing power of atlas 900 can be widely used in scientific research and business innovation. For example, astronomical exploration, oil exploration and other fields require huge data calculation and processing. It may take several months to work, but now it's just a matter of seconds for Atlas 900

next, let's share a case of astronomical exploration, which was jointly created by Huawei, Shanghai Observatory and ska. Astronomical research is highly dependent on massive data analysis and calculation, and cannot be separated from strong computing support

the following is a picture of the southern hemisphere

Copyright © 2011 JIN SHI