☞ This is the 15907th article published by Metal Processing (mw1950pub)
Editor’s Note
In the context of the integrated development of technologies such as IoT, 5G, AI, and big data, smart edge computing has rapidly transitioned from concept to implementation. Especially since the outbreak of the COVID-19 pandemic, an increasing number of network applications, such as video surveillance, smart retail, smart manufacturing, and smart cities, have rapidly developed, leading to more data being stored and processed at the edge. According to Gartner’s predictions, by 2025, 75% of data will be generated outside data centers at the edge, indicating that digital transformation is moving from the cloud to networks and endpoints. However, smart edges also pose new challenges for industry enterprises. How will technological trends develop? How can end-to-end digital infrastructure be built? How can related applications be implemented to empower industries like healthcare, agriculture, smart transportation, and retail?
According to Zhang Yu, the global chief technology officer of Intel’s Video Division, the increasing importance placed on smart edge computing is due to the profound digital transformation that society is undergoing. In this process, the operations of various industries must rely on a complete set of end-to-end digital infrastructure to function normally. The edge is an organic component of the entire system, and its role is becoming increasingly prominent.
“The emphasis on ‘end-to-end’ is because the services required by this system are all end-to-end services, from data collection at the front end to data transmission, and then to data processing, storage, and feedback at critical nodes. The entire service is a complete end-to-end process, and only such a complete process can be effectively accomplished,” Zhang pointed out. In such a data transmission and processing process, some data can be completed at central nodes, while others must be processed at edge nodes. This is determined by the different needs of data applications. Applications that do not require real-time processing can transmit data from the application front end to the remote cloud, where there is more storage space and relatively abundant computing power, allowing for computation and storage at lower costs. However, for applications requiring high real-time performance, such as autonomous driving and industrial internet, it is not suitable to transmit data to remote locations for processing. Such computations often need to be supported by edge computing architectures like computing power networks. This is the main reason why smart edge computing is increasingly valued.
“Without the intelligence of IoT and the ubiquity of computing power networks, it is impossible to support the real-time and efficient computing needs of applications at the front end,” Zhang emphasized.
For smart edge computing, three development trends are anticipated. First, the construction of digital infrastructure will be a collaborative system integrating “end-edge-cloud.” None of the three can be absent, and their roles differ based on application scenarios, with the edge sometimes playing a more important role in data processing, while at other times the cloud plays a more significant role.
Second, the role of software in digital transformation cannot be overlooked. This is often referred to as “software-defined.” With the development of remote work, the demand for flexible networking has been increasing. This requires a network that can be flexibly deployed, not only in the core network but also in the configuration of data centers, which must become increasingly flexible. Data centers must be built in a programmable manner, with programmable Ethernet chips and programming languages becoming crucial supports for data center construction.
Third, the penetration of artificial intelligence in edge computing is becoming increasingly widespread. Although the current wave of AI applications began in data centers, where a large amount of data processing was completed in the cloud during the initial phase, as technology develops, more and more data processing is shifting to the edge. The integration of edge computing and artificial intelligence is inevitable.
In summary, against the backdrop of the integrated development of technologies such as IoT, 5G, AI, and big data, smart edge computing is rapidly transitioning from concept to implementation. According to IDC, by 2025, approximately 75% of data will be generated and processed at the edge. The massive data generation is continuously increasing the demand for edge computing power. Smart edges will become an important computing domain for end-to-end digital infrastructure in the digital age.
In the rapid development of smart edge computing, chips play a key role. “In fact, innovative applications of smart IoT must be realized through underlying chips and supporting software. From the perspective of chips, as the demand for computing power grows, the requirements for chips will also increase,” Zhang pointed out.
Intel has long been deeply involved in the IoT industry, providing a diversified product line to meet user needs. At this year’s International Consumer Electronics Show (CES 2022), Intel launched its 12th generation Core processors (code-named Alder Lake S series and H series). This is Intel’s first processor series optimized for the edge, using a high-performance hybrid architecture that integrates performance cores, efficiency cores, and hardware thread schedulers (Thread Director) to optimize the acceleration of IoT application innovation, suitable for users in retail, manufacturing, healthcare, and video sectors for graphics, multimedia, display, and AI computations.
Zhang also emphasized that while Intel is committed to improving the performance of its chip products, it also places great importance on achieving sustainable green computing concepts. “We cannot solely emphasize the increase in computing power; we must also recognize the importance of improving efficiency ratios. Intel always focuses on promoting the orderly development of society in a sustainable manner.”
It is reported that from 2010 to 2020, the energy efficiency ratio of Intel’s Core product line improved by 14 times. From 2020 to 2030, Intel’s goal is to further improve the average energy efficiency ratio of its product line by 10 times on the existing basis. To achieve this, Intel has introduced the concept of “big and small cores” in its 12th generation Core processors, including performance and efficiency cores, utilizing different cores to match computing needs for optimal load and reduce overall power consumption.
Heterogeneous computing is also an important technological direction for addressing differentiated demands and improving processor energy efficiency ratios. Many loads have typical characteristics, whether in video conferencing scenarios, smart retail scenarios, or cloud gaming scenarios, the underlying technology involves video encoding, decoding, and transcoding. For these relatively fixed loads, using dedicated hardware for processing is more efficient and consumes less power. Therefore, Intel has long integrated integrated graphics into its Core product line, forming a hardware unit dedicated to video encoding and decoding. Of course, for users with higher performance requirements, Intel can also provide solutions with discrete graphics cards. The overall approach is to improve energy efficiency ratios through heterogeneous architecture.
How to leverage these hardware products relies on software support. Zhang emphasized that a good software tool can enable developers to fully utilize hardware performance. Intel has been working with partners in this area, including launching oneAPI to realize access to heterogeneous hardware resources through the underlying DPC++ open interface. For developers, using oneAPI allows unified programming across different hardware.
Intel also launched the OpenVINO toolkit to accelerate AI inference at the edge. OpenVINO enables developers to train within different AI frameworks. In some AI model scenarios, OpenVINO can provide up to 7 times inference acceleration, effectively improving work efficiency and increasing platform competitiveness.
With the development of smart edge computing, its successful application has been realized in several areas.
Industrial manufacturing is one of the key areas of penetration and development for smart edge computing. A recent research report by MarketsandMarkets shows that the global smart manufacturing market size was $88.7 billion in 2021 and is expected to reach $228.2 billion by 2027, with a compound annual growth rate of 18.5%. In this regard, Zhang Yu stated: “Machine vision, as an important technology for the intelligent transformation of the manufacturing industry, is evolving rapidly with the high-speed development of artificial intelligence, and its market size is growing rapidly. In fact, the current wave of artificial intelligence explosion is based on deep learning based on image recognition. Currently, about 80% of the data collected in the IoT field is related to graphics and images.”
Based on this, Intel recently partnered with Xunbu Technology to launch a series of machine vision development kits. The development kits are framed around the Industrial Edge Insight Platform (EII) and visual software optimization package, integrating software tools and libraries such as OpenVINO, DPC++/C++ compiler, oneAPI Math Kernel Library (oneMKL), Vtune™ Profiler, IPP, OpenMP, and TBB, along with preset typical industrial application scenarios and algorithm optimization reference methods, providing users with an end-to-end software framework for developing machine vision applications. Currently, the development kits have been successfully applied in industries such as automotive manufacturing, 3C/semiconductors, food packaging, and logistics warehousing, effectively assisting the manufacturing industry in its evolution toward intelligence and informatization.
The retail market is becoming increasingly competitive, and the significant value demonstrated by artificial intelligence is gaining widespread attention from global retailers. Through the application of artificial intelligence technology, retailers can gain a better understanding of consumer preferences, provide personalized and unique services to consumers, and enhance the attractiveness of retail services. At the same time, artificial intelligence also helps retailers automate more processes, resulting in increased profits. Data shows that by 2035, artificial intelligence will increase profit margins in retail and wholesale businesses by nearly 60%.
In response to this emerging application market, Intel, along with Hanshuo Technology and Microsoft, has jointly launched smart edge solutions for the retail industry, primarily applied in smart shelf management and self-checkout loss prevention. “People, goods, and places” are the main elements of traditional retail; this solution integrates Intel’s full-stack technology from software to hardware, enabling retail customers to build high-performance and easy-to-manage smart retail management systems.
In addition, edge intelligence has broad application potential in smart healthcare, smart transportation, and accelerating digital transformation for enterprises. “The development space for artificial intelligence in edge networks is vast, and market demand is very clear. Previously, people thought artificial intelligence was a high-end technology with a high barrier to entry and difficult to deploy. However, the reality is that more and more companies are starting to deploy and operate artificial intelligence technologies and achieving success. This situation precisely indicates that the barriers to using artificial intelligence are lowering, and it is also a typical example of the ubiquity of artificial intelligence,” Zhang pointed out. In this process, Intel’s technologies, including the OpenVINO toolkit and a series of hardware products, have also played a key role.
In the process of accelerating the implementation of smart edge computing, how to better adapt it to local needs and overcome the “last mile” bottleneck is also a very critical aspect. During the interview, Zhang particularly pointed out: “The way technology is implemented in data centers is different from that at the edge.” After all, the platforms, computing power, and conditions are different, so the implementation methods will not be the same. This means that when artificial intelligence is implemented at the edge, it is necessary to solve specific problems that only arise at the edge.
For example, artificial intelligence is generally divided into two major stages: the training stage and the inference stage. The training stage requires the use of a large amount of data for training, and before training, the data must be labeled to highlight the objects of interest, and then the labeled images are transmitted to the training platform for training, ultimately generating a usable artificial intelligence network model for inference.
However, this model may encounter challenges at the edge. In reality, the edge often does not have enough data available for training. For instance, defect detection is one of the most common applications of artificial intelligence in smart factories. However, the probability of defects occurring on a normal production line is not very high. This means that the sample data available for training is relatively limited. How to train a usable model with a small sample is a problem that needs to be addressed in smart edge computing.
Moreover, during training, labeling is required, but in edge applications, the personnel operating artificial intelligence are often the production staff on the production line. These individuals find it challenging to have extra energy for labeling work. Therefore, developing automated labeling tools to reduce the burden on operators and ensure that the closed loop of training, labeling, and inference can truly operate is also a problem that Intel needs to solve when implementing artificial intelligence at the edge.
“In fact, many problems at the edge need to be solved in a specific manner,” Zhang emphasized.
Of course, in the field of smart edge, all of Intel’s work relies on the support of partners. “Intel has long been deeply engaged in the IoT industry, providing a rich variety of product types. We hope that through these different combinations of products, partners can have greater room for development in the IoT field,” Zhang pointed out.
When discussing Intel’s competitive advantages in the smart edge field, Zhang attributed it to the extensive partner base. “This allows us to better understand user demands and formulate software and hardware solutions based on such demands. Whether these users are located at the network front end, edge, or data center, Intel can provide end-to-end solutions that align with the development trends of digital infrastructure.”
Although IoT has infinite possibilities, the issue of fragmentation is always a major challenge for suppliers. In this regard, Zhang stated that on the one hand, we must do our own products well, making them as universal as possible; but in terms of project implementation, we will always rely on close cooperation across the entire industry chain.
Different markets have different demands. The characteristics of the Chinese IoT market are rapid development and many innovative customers. These customers may not be large in scale, but they have strong innovative capabilities. In recent years, Intel has worked extensively with domestic partners, including system integrators, to integrate product technologies across different segments of the industry chain to address fragmentation issues.
In 2016, Intel, along with partners, jointly established the Edge Computing Industry Alliance. This is currently the largest alliance in China focused on edge computing, with over 300 member units, covering chip manufacturers, system integrators, software developers, ODMs, OEMs, and other manufacturers. Through this alliance platform, solutions for IoT and edge computing, as well as industry solutions, are discussed collaboratively to address fragmentation issues.
“Water nourishes everything without contention” might be the best description of Intel’s positioning in the entire ecosystem. Over the years, through quiet and diligent work, Intel has connected various links in the edge intelligence industry chain, including original design manufacturers (ODMs), original equipment manufacturers (OEMs), system integrators (SIs), and independent software developers (ISVs), providing an end-to-end overall solution oriented towards digital infrastructure.
Intel CEO Pat Gelsinger proposed the concept of four super technological forces at the “2022 Intel On Industry Innovation Summit” this year, including ubiquitous computing, ubiquitous connectivity, artificial intelligence, and infrastructure from cloud to edge. Zhang emphasized that these four super technological forces are essentially the four pillars required to build a green, intelligent, end-to-end digital infrastructure in the future. If these four super technological forces are used comprehensively, what will be constructed is an end-to-end digital infrastructure. In the future, this digital infrastructure will support more new creations and applications. These infrastructures have two characteristics: one is intelligence, and the other is greenness.
-End-
☞ Source: China Electronics News ☞ Responsibility Editor: Shao Yujie ☞ Reviewer：Wu Xiaolan ☞ Media Cooperation: 010-88379790-801 ☞ The only submission website for Metal Processing magazine:http://tougao.mw1950.cn/