“Expected high growth, did not expect blowout” domestic chips meet the golden development period

In 2020, the “new infrastructure” included in the government work report will mainly focus on digitalization and intelligence. Focusing on 5g, cloud computing and artificial intelligence, we will spare no effort to build information digital infrastructure and boost the high-quality development of China’s social economy. “New infrastructure” emphasizes the important role of digital infrastructure in the development of high-tech. < / P > < p > the form of data center is constantly evolving. In April 2020, Inspur proposed that the future data center will evolve into “Intelligent Computing Center”. How to understand the concept of “Intelligent Computing Center”? < / P > < p > the application and operation mode of data center in each historical period are different. Inspur believes that the future data center will take “cloud computing + big data + artificial intelligence” as the core feature, and puts forward the concept of “Intelligent Computing Center”. The proposal of Intelligent Computing Center has considered both technology and infrastructure. < / P > < p > the Intelligent Computing Center is the most important computing power production center and supply center in the smart era. It takes the converged architecture computing system as the platform, takes data as the resource, and can drive AI model with powerful computing power to carry out deep processing of data, continuously produces various intelligent computing services, and supplies them to organizations and individuals in the form of cloud services through the network. < / P > < p > first of all, from the technical level, the Intelligent Computing Center will be more oriented to artificial intelligence, based on cloud computing, through processing massive data, outputting various AI services to the outside, so as to achieve the integration of “cloud + number + intelligence”. At present, Artificial Intelligence Computing demand is growing exponentially, and will occupy more than 80% of the computing demand in the future. Therefore, we propose “intelligent computing”. Compared with the traditional Supercomputing Center or commercial data center, “intelligent” emphasizes its important significance for the development of artificial intelligence. < p > < p > secondly, the Intelligent Computing Center is a public infrastructure to provide public services for various industries. Now the data center of the Internet industry is developing rapidly, and some innovative businesses and technologies have developed rapidly. In contrast, many traditional industries still have a long way to go in terms of digital transformation. One of the important obstacles is the lack of digital infrastructure. Therefore, Inspur believes that the intelligent computing center should have the public nature of infrastructure, just like the mobile network and highway, so as to achieve universality and give full play to social value. < / P > < p > in the future, computing power is productivity. Inspur puts forward the “Intelligent Computing Center”, which is committed to building it into a computing power production center and supply center, providing intelligent computing services to organizations and individuals in the form of cloud services in an efficient and HP way. < / P > < p > “integration” is becoming the way of data center construction. The tide also mentioned “integration of Intelligent Computing Center”. How to understand “integration of Intelligent Computing Center”. The first level is the technology integration within the Intelligent Computing Center, namely “integration architecture”. Inspur thinks that in the future data center, the boundary between computing, storage and network will be more and more blurred, and they will be merged into “resource pool”, and resources will be divided by software. < / P > < p > the second level of integration is the integration between the intelligent computing center and the Intelligent Computing Center. No matter how large a data center is, it is impossible to solve the problems in China. Therefore, the future data center will be physically decentralized and logically centralized. This requires us to realize interconnection at the standard level at the beginning of the construction of Intelligent Computing Center. In the future, different intelligent computing centers can also realize the flow of data and business. The third level of integration is the integration of Intelligent Computing Center and various industries. If the intelligent computing center wants to become an infrastructure, it must be deeply embedded in various industries, and fully integrate with CT, OT, intelligent manufacturing and traditional industries. Only in this way can the intelligent computing center cover all scenes of production and life. In the future, computing power will become an important resource. However, Moore’s law rate is gradually reaching the ceiling. What impact will this have on the computing power supply of the Intelligent Computing Center? < / P > < p > the speed of the new chip technology update iteration has slowed down, and relying solely on the chip to improve the computing power performance has encountered a bottleneck in the development. Although Moore’s law is slowing down, there are still many opportunities to improve the computing power. < / P > < p > first of all, the rise of dedicated computing chips and the research and innovation of new computing modes provide a new solution to alleviate the demand for computing power in the case of slowing down of traditional silicon-based semiconductor technology. The use of special chips to make up for the lack of computing capacity has been widely used. In recent years, the rise of AI chips and acceleration chips, the innovation and use of GPU, FPGA, XPU and other accelerator chips, and the application scenarios such as AI deep learning training and online reasoning can well meet the demand of AI computing power explosive growth. < / P > < p > Second, some new computing technologies, new storage media, new transistor design methods and molecular level computing technologies are also accelerating the application, such as quantum computing, biological computing, photon computing, etc., which provide more space for accelerating innovation at the computational level; another example is non-volatile memory dcpmm, which has a maximum single memory capacity of 512gb, As a result, the reasoning activity density is doubled, and the cost per unit activity is reduced by nearly 50%. < / P > < p > Third, at the level of architecture, the rapid development of fusion architecture is also accelerating the collaborative innovation of software and hardware design. At the hardware level, resource pooling is realized through hardware reconfiguration. CPU and GPU, FPGA, XPU and other accelerators will be more closely combined, and the integration of heterogeneous computing chips will be realized by using new ultra-high-speed internal and external interconnection technologies such as fully interconnected nvswitch, CXL and open capi. At the software level, based on the reconfigurable hardware resource pool and flexible organization, different resource pools will be formed into professional server, storage and network systems, And realize the efficient management and scheduling of resources and the flexible flow of data in the pool resources. When AI is combined with software definition, it gives software definition more advanced meaning. < / P > < p > at present, Inspur’s whole machine solution is also constantly innovating its architecture. Inspur drives the innovation of “converged architecture” through two levels of “hardware reconfiguration” and “software definition”. Under the guidance of the integration architecture technology route, Inspur will continue to consolidate the advantages of computing power system and cloud digital intelligence platform, provide artificial intelligence server, general computing server and key computing server, and also provide storage and network products for massive data and high-speed interconnection. In addition, the technology stack continues to extend to the bottom, facing different scenarios, and finally forms a leading product system covering chips, systems, platforms and algorithms. < p > < p > Inspur is an IT enterprise with a long history in China, which is also constantly pursuing technological innovation. In what ways does Inspur respond to the development of “new infrastructure”. Open computing is the future industry trend and the cornerstone of Intelligent Computing Center. Regardless of the underlying chip, system, platform, algorithm, the development of the whole technology and product system tends to be more open. In the process of promoting the development of open computing, Inspur has participated in risc-v, openstack, OCP, odcc, open19 and other mainstream open software and hardware communities and organizations. At the same time, taking openstack, k8s and other open infrastructure as the base, and integrating aiops and artificial intelligence training and reasoning platform, the ability of cloud digital intelligence integrated open platform is constructed. Open computing has become a very important direction in the whole business strategy of Inspur. < p > < p > secondly, Inspur also took the lead in the layout of artificial intelligence. At the level of artificial intelligence, Inspur builds AI capabilities from three aspects of products, platforms and ecology. At the product level, Inspur has built the world’s leading AI server; at the platform level, it has launched aistation, a leading AI resource management platform, to achieve efficient scheduling and management of AI resources. In addition, in order to promote the ecological construction of artificial intelligence and accelerate the AI application of industry, Inspur proposed “meta brain ecology”, which combines left-hand partners with AI development core competence and right-hand partners with overall industry solution delivery ability to truly solve the AI landing problem in traditional industries. < / P > < p > finally, Inspur provides customized products and services for large CSP users and traditional industry users through the innovative JDM collaborative innovation mode. Inspur has a leading rapid R & D capability, and can design products and solutions that meet the needs of users in the shortest time. At present, starting from 0, Inspur can provide prototype within 3 months, and has the capacity of batch supply in 9 months. On January 10, 2019, Inspur created a delivery record of 10000 node servers per day delivered by a single data center in Baidu data center. These servers are in the form of a whole cabinet and can be delivered in a modular way. Xiaomi new machine real machine exposure center dig hole screen design or high color thousand yuan machine