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

A few days ago, the International Intelligent computer Conference (BenchCouncil 2019) hosted by dawning of Chinese Science and Technology was held in Shenzhen. With the popularity of AI, the concept of intelligent computing is becoming more and more popular. At present, the prying power of AI to social and economic productive forces is closely related to computing power. It is precisely because of the strong computing power that we can deal with the huge amount of data in the industrial scene and improve the industrial efficiency; it is also because of the computing power that the comprehensive digitization of cars, roads and people in the travel scene can be targeted to promote the development of self-driving.

Therefore, high-performance computing, which represents the most powerful standard of computing power (hereinafter referred to as HPC), is more and more closely related to AI. Today we can try to explore intelligent computing and the future it shapes.

Clever and fast: why isn't traditional computing the ultimate destination of AI?

In his speech, Li Guojie, academician of the Chinese Academy of Engineering, chief scientist of the Computing Institute of the Chinese Academy of Sciences, and chairman of the dawning Company, pointed out that artificial intelligence and intelligent computers have been the direction of the Computing Institute of the Chinese Academy of Sciences for decades.

Today, the high degree of fit between HPC and AI is no accident. Today, the development of this wave of AI is closely related to the development of deep neural network. The number of layers of neural network is more and more, the amount of data carried is larger and larger, and the requirements for computing power will continue to improve. So we can see that the "Depu AI" Libratus developed by Carnegie Mellon University uses the supercomputing ability to expose, disassemble, input and calculate the rules of chess and cards; Nvidia, Intel and other manufacturers have also launched suitable supercomputing systems for scenarios such as autopilot AI training.

However, Academician Li Guojie told us that although overcalculation is currently an indispensable infrastructure for AI, the traditional overcalculation does not fully meet the needs of AI.

"supercomputing is a computer that calculates fast, and intelligent applications such as deep learning need to be calculated very skillfully."

In fact, the improvement of computing power itself is on the one hand, on the other hand, we need to consider how to make intelligent applications make better use of computing power, and both sides reasonably improve the comprehensive efficiency of intelligent computing. Such as cloud data centers, intelligent workstations, intelligent terminals and smart devices for human-computer interaction, as well as cloud-based intelligent computing and intelligent application-oriented supercomputers, which adapt to the needs of AI computing.

Academician Li Guojie mentioned that for AI applications, computing devices can be adjusted in terms of hardware structure, and the more complex the structure, the simpler the algorithm can be applied. In the future, artificial intelligence needs not only computing power, but also complex hardware structure, which may invent new intelligent computers, but at least in the last 20 years, intelligent super is a research direction that should be paid great attention to.

Intelligent grip under the three engines: dawning's Intelligent Computing solution

If intelligence is the future of computing, as a worldwide leader in computing, how will Dawn of Chinese Science promote the transformation from traditional computing to intelligent technology?

Song Huaiming, chief scientist of big data and director of big data Intelligence Research Institute, mentioned that the layout of dawning is "strong computing power, universal computing and full-stack intelligent computing scenario" to develop intelligent computing for the three major engines. to provide users with "convenient, economical and comprehensive" intelligent computing resources.

In terms of "strong computing power", dawning has built an intelligent computing product line for deep learning, machine learning, etc., and different types of chips suitable for different scenarios, such as finance, image, education, and so on, such as GPU, which has advantages in image computing. And FPGA, which has more advantages in heterogeneous computing, have launched applicable products, such as intelligent supercomputing, deep learning server, on-demand storage, super-converged all-in-one machine and so on.

Through the adjustment of product structure, it can meet the computing needs of users in different scenarios, thus giving better play to the efficiency of AI application computing power.

The so-called "universal computing" is to service the computing power from a "heavy asset" to a resource readily available to users and customers. In recent years, the dawn of Chinese science has gradually changed people's perception of supercomputing-supercomputing is not just a pile of servers that occupy a huge area, allowing users to use "hard drives" to install programs to run on supercalculations. it can also be light, convenient and readily available.

So we can see that China Science Dawn has launched an artificial intelligence service platform, an advanced computing service platform known as "supercomputing OA", as well as cloud data centers, advanced computing centers all over the country, and so on. The advanced computing power is packaged into cloud, integrated hardware and other forms, so that users can obtain computing resources through cloud computing, edge computing, general computing and other ways. So that users do not need to make a big fight, build a huge computing center, but also get computing power in intelligent applications.

Finally, dawning of China Science and Technology also launched the "full-stack intelligent computing scenario" to promote the intelligent solution of the whole scene of the industry. To put it simply, intelligent computing is a starting point and a starting point, through which the dawn of Chinese science touches the industry and implants integrated computing and algorithm solutions in many fields, such as security, medical care, scientific research, Internet, finance, environment, agriculture, manufacturing, and so on. To help enterprises achieve intelligent business transformation in an ecological way.

On the eve of the revolution: conceptual upgrading from computing to intelligent computing

To put it simply, the driving engine of dawning for the transformation from traditional computing to intelligent computing can be divided into three levels.

The first layer is the change of the design of the computing product itself, which constantly drives the computing product to be close to the application mode of AI computing, and the second layer is the change of the survival state of the computing product, so that the HPC is no longer just piled up in the server, but to the terminal and the cloud, reducing the comprehensive cost of the application. The third layer is the change of computing ecology, so that computing is not only a "resource", but also a combination of other intelligent services to create a channel for the intelligent transformation of enterprise business.

From these three levels, it is not difficult to find that the transformation from traditional computing to intelligent computing not only means the change of technology, but also the upgrading of concepts.

The traditional concept of "computing" often exists as a power driver, and the computing side passively provides what resources are required from the application point of view. Under this concept, it will make the development of computing technology more isolated, and it is difficult to break through the limitations of the existing application demand, which may lead to the development of computing technology but can not be transformed into the overall socio-economic capacity improvement.

However, in the development of HPC and AI, we have seen that computing power is like soil. Only when the soil is rich, can we make progress in algorithm technology. The development of computing power and algorithm ability must be intertwined spirally forward. The algorithm determines the geometry of the value of the computing power, and the computing power determines the upper limit of the research and development of the algorithm.

However, with the Moore's Law that the number of components that can be contained in integrated circuits doubles every 18-24 months, quantum computing, two-dimensional materials and other technologies continue to make breakthroughs. What we are about to witness is a computing revolution that rewrites the status quo. The Cambrian explosion of brand-new computer architecture proposed by Academician Li Guojie is one of the events doomed to happen in this revolution.

Furthermore, the transformation of intelligent computing proposed by dawning of China Science and Technology has mapped out the direction for the upcoming computing revolution: first, to better meet the needs of AI for computing power through product innovation, then to help AI users obtain numeracy more easily through service innovation, and finally to foster the continued growth of AI's demand for numeracy through ecological innovation.

When Kilby developed the world's first integrated circuit sixty years ago, I believe no one imagined that computers and the Internet would work together to shape our world into what it is today. Similarly, the intelligent computing revolution we experience today will also lead the world to a more beautiful and unknown "Cambrian".

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.