Turning technologies such as artificial intelligence, 5g and intelligent edge are at an unprecedented historical intersection. They are accelerating their breakthrough and integration. They are becoming a new infrastructure in the intelligent world and an important driving factor for the future growth of traditional chip giants such as Intel.
At the cloud summit of the 2020 world Artificial Intelligence Conference (waic), song Jiqiang, President of Intel China Research Institute, pointed out in his speech: “we have entered the intelligent era. The ‘intelligent x effect’ is a key opportunity. 2020 is an important time node to promote the intelligent transformation of all walks of life.”
“Intel’s ambition for AI is unprecedented, and the company has great potential for development in the future.” song Jiqiang said that artificial intelligence and data analysis are decisive workloads in the next 10 years, key turning technologies, and important drivers for Intel’s future growth.
“The only AI capable chip company”
“The technology supporting artificial intelligence is not only computing, but also storage and communication. Intel is a provider of end-to-end full stack solutions for artificial intelligence chips,” said Zhang Yu, chief technology officer and chief engineer of Intel’s Internet of things division in China, Intel is the only semiconductor company that can fully provide solutions from computing to storage to communication.
Zhang Yu, chief technology officer and chief engineer of Internet of things division of Intel China
In Intel’s AI chip “arsenal”, Zhang Yu introduced that in addition to the latest third-generation Xeon processor just released in June this year, it integrates new instruction sets related to AI, which can accelerate the performance of artificial intelligence in training and reasoning; At the edge, the power consumption of Intel’s second-generation movidius chip is only about 2W, but it can provide up to 1t computing power. It can comprehensively optimize the convolution layer, full connection layer and activation function in the deep convolution network.
“So movidius chips often appear in smart products such as smart cameras and translation pens.” Zhang Yu revealed that the new generation of movidius chips will be on the market soon.
According to Zhang Yu, in addition to computing, Intel also has new products in communication and storage. For example, in February this year, Intel released SOC chip based on IA architecture, which can be used to provide 5g base station solution; The new generation of aoten memory released in June has increased the bandwidth by 25%.
Intel takes data as the center to build the comprehensive capabilities of intelligent connection, intelligent storage and intelligent computing. Intel takes the six technical pillars (process and packaging, XPU architecture, memory and storage, interconnection, security and software) as the engine to promote intelligent innovation, integrates artificial intelligence into all aspects, and combines software and hardware with the comprehensive product layout of CPU + GPU + FPGA + ASIC, Help customers realize intelligent deployment from cloud to end.
“Intel has put forward the concept of XPU to deal with a wide range of applications. Intel’s goal is not to use one architecture or one processor product to realize a variety of applications, but to use a series of architectures and different models of products to solve it,” song Jiqiang said.
AI 3.0 does three things right
“Judging from the current development trend of the whole academic and industrial circles, AI is moving from stage 2.0 to stage 3.0. Stage 2.0 is data-driven and trains deep learning models through a large number of taggable data to help us complete tasks in some specific fields.” song Jiqiang believes that AI version 3.0 will become more general, It can be used in many fields. In the face of the artificial intelligence 3.0 stage of carrying forward the past and forging ahead, as long as you have to do three things right.
First, let AI be interpretable. Artificial intelligence should not only complete tasks, but also know why it is right and where it is wrong, find the reasons for correction and promote steady improvement. This is very important for medical treatment, safety, transportation and other fields.
Second, we should improve the ability of artificial intelligence to use a small amount of data for continuous learning. In the future, it is impossible to label a large amount of data generated in many application fields of artificial intelligence, and some data cannot be completely collected. Some data are visible and some data are invisible. It depends on knowledge and reasoning, which puts forward requirements for the learning ability of artificial intelligence. Using a small amount of data can slowly build a complete cognition.
Third, the robustness of artificial intelligence should also be improved. AI with high robustness can eliminate wrong data.
“The whole industry is gradually moving towards 3.0. For example, the neural mimicry computing done by Intel Research Institute is a gradual process that allows machines to learn by themselves, and the learned model can be explained and is relatively robust. At the same time, it further improves the energy efficiency and does not need a lot of electricity to train a model.” song Jiqiang added, At present, the industry is migrating to phase 3.0, but at the same time, the technology of phase 2.0 of artificial intelligence is also being commercialized on a large scale.
Full stack R & D behind full stack products
Accelerate the integration of “Ai, 5g and intelligent edge”. In Song Jiqiang’s view, smart future city is an ideal carrier for the implementation of turning technology applications. For example, Nanjing “future science and Technology Intelligence Center” built by Intel, Nanjing Economic and Technological Development Zone and a number of ecological partners cultivates innovation ecology and constantly explores the future development of smart city by jointly building 5g + smart park.
Zhang Yu believes that the industrial Internet is essentially an end-to-end system of “edge cloud collaboration”. Among them, about 50% of the data processing occurs on the edge side, and the other 50% occurs in the cloud. A large part of these data need to be processed by artificial intelligence technology. Intel’s advantage is that it can help partners build cloud to end artificial intelligence systems.
According to statistics, the number of developers on IA architecture exceeds 20 million. For this large group, Intel has created projects such as Intel AI practice day, Intel AI builders, Intel AI Developer program, openvino Chinese community, etc.
“After the hardware is integrated, it is also very important for software personnel to make good use of it and release its ability.” song Jiqiang believes that the ability of hardware is fundamental, but how to make good use of it depends entirely on software, “With oneapi as a large framework, open source is oriented to the whole industry. There are many different performance libraries. For example, for visual computing, Intel has openvino performance library. The feedback from the industry is very good.”
In fact, Intel is not only a technology provider, but also a technology practitioner. For example, in Intel Dalian factory, it is necessary to detect wafers in real time and accurately to ensure the yield of products. Through the artificial intelligence software and hardware technology based on Intel, the detection efficiency is improved by 100 times compared with the pure manual detection method.
“Intel sees not only the present, but also the next 3 to 5 years, and even some opportunities that can bring major breakthroughs to the AI field in the next 8 or 10 years, such as quantum computing.” song Jiqiang thinks this shows Intel’s deep foundation and strength beyond the whole stack.
Editor in charge: PJ