Edge AI has a big trick! AI models support virtual digital humans and robotic arms, Intel's edge platform helps partners accelerate innovation

(Fan report by Zhang Ying) On July 25, at the 17th Edge Computing Industry Conference, the edge virtual person "Xiaoying" attracted the attention of the participants. "Xiaoying" is a 3D virtual digital person based on Intel Core and Intel ARC770 technology. In addition to achieving 4K 60-frame high-definition rendering, it also integrates a large language model and RAG technology locally, and can be fully deployed on the edge side. What can it do? Xiaoying's answer is that he can be competent for many aspects of work, including serving as a customer service representative in a business hall, a shopping guide in a shopping mall, or a guide in a museum, bringing vivid and interesting historical and cultural explanations to visitors.

This is a vivid example of the implementation of edge AI applications. At this conference, Intel and more than 400 ecosystem partners and customer representatives gathered together to discuss the future development trends of edge AI and introduced many wonderful examples based on Intel edge AI solutions in the fields of education, manufacturing, networking and smart retail.

What are the market trends of generative AI? What are the latest insights into edge applications? What innovative achievements have Intel and its partners demonstrated? Sachin Kat, Intel senior vice president and general manager of the Network and Edge Division, Dan Rodriguez, Intel vice president and general manager of the Network and Edge Solutions Division, and Dr. Wei Chen, Intel vice president and general manager of the Network and Edge Division in China, shared their wonderful views and innovative cases, which are reported in detail in this article.

The penetration rate of edge AI will reach 50% in 2026, and China's generative AIGC will develop rapidly

Wang Zhicong, vice president of Intel's marketing group and general manager of Intel China, said that Gartner data shows that the market size will exceed US$500 billion for the first time in 2021, and semiconductor technology is currently driving the transformation of almost all industries. With the endless growth in demand for computing power, the semiconductor market size is expected to double to US$1 trillion by 2030. The main driving force behind this is still the digital transformation of various industries around the world and the new opportunities for integration.

He emphasized that as the foundation of the real economy, China's manufacturing industry continues to expand in scale, its share steadily increases, its structure continues to optimize, and its investment continues to grow. More and more companies are realizing digital transformation through AI, promoting high-quality industrial development with new productivity. In addition to the in-depth development of edge computing in various industries, such as retail, manufacturing, finance, healthcare and infrastructure, we also observe that there is still great potential for the application of edge computing technology in networks and edge computing.

The size of China's generative artificial intelligence market will reach US$3.3 billion this year and will rapidly double in the next few years, with broad market development prospects. Intel will focus on its superior resources to accelerate the promotion of AI technologies represented by large models to help traditional manufacturing industries transform and upgrade, while providing enterprises with open, plug-and-play solutions and building an open, scalable system architecture to help enterprises win the future.

Three steps of AI evolution: Intel's edge AI platform solution helps customers

Sachin Katti, senior vice president and general manager of Intel's Network and Edge Group, shared his insights on the blueprint for AI evolution. He believes that AI is undoubtedly expected to become the most disruptive technology in recent decades, and the AI ​​transformation of enterprises will be a gradual evolutionary process. The first is the AI-assisted era. At this stage, AI serves as an auxiliary tool or assistant for humans, enhancing human capabilities and improving work efficiency. Then, humans will soon be inferior to the AI ​​assistant era. Enterprises will develop models for specific fields, and by allowing AI agents to autonomously handle the workflow in that field, humans will eventually usher in the era of fully functional AI.

Sachin Katti said: "Intel is focused on helping enterprises simplify the complex process of deploying AI on PCs, edges and data to realize the vision of 'AI everywhere'. Intel has a comprehensive and broad foundation. From Atom to Xeon, we have launched the Atom x7000RE/x7000C series, Core and Core Ultra series, Xeon 6700E series, Arc for the edge and IPU E2100 network adapter. Through software-defined simplicity and an open and diverse ecosystem, Intel can bring rich choices to customers."

Dan Rodriguez, Intel vice president and general manager of the Network and Edge Solutions Division, said: "Intel has been in the field of edge computing for many years and has supported a wide range of industries, including smart cities, networks, etc. Our products can meet a wide range of needs through edge computing."

He pointed out that Intel Core processors are not only high-tech, but also integrate graphics processing capabilities and AI functions. This allows every application case to take full advantage of the multi-purpose processor (MPU). The design of Core products is excellent, combining GPU and customer needs, as well as large model application needs. Customers can use and GPU, combined with Intel's OpenVINO tool kit, to run AI applications on the platform more easily. We provide platform-level solutions to help customers better manage their devices in edge systems. In the entire ecosystem, in addition to chips and software, AI kits are also included. These AI kits have a variety of applications, enabling participants in vertical fields to use better AI tools to create better and more powerful products.

Under the AI+ trend, Intel will focus its technological advantages on promoting the implementation of scenarios and industry applications

Dr. Wei Chen, Vice President of Intel Corporation and General Manager of the Network and Edge Business Unit in China, said: "In the advancement of the entire intelligent edge and AI, application scenarios are a dominant direction, which guides how our technology develops."

He analyzed: Intel's support covers two major levels. First, it provides open technical support. We have in-depth cooperation with many partners, and these cooperations have lasted for more than 10 or even 20 years. Our investment and output in technology play a very important role in the upgrading of the entire industry ecosystem. Second, we focus on solutions and continue to deepen our industry. Facing the arrival of AI+, we have adopted three main strategies: First, based on existing intelligent solutions, we integrate AI technology into existing intelligent solutions; second, we actively embrace the current explosive growth of large models and generative AI technologies. Today, we also demonstrated how our ecosystem can quickly adapt to new changes through actual application cases; third, support for the entire ecosystem. The entire strategy is based on application scenarios, supplemented by technology, to jointly welcome the new era of AI."

Chen Wei specifically mentioned Intel's Tiber edge platform, which is based on the service level experience accumulated in vertical industries over the past decade, integrating them into a platform. Dan Rodriguez today specifically introduced software tools for vertical industries, such as AI Speech f Vision, AI Speech for Factory, etc. The development of these tools actually benefited from the close cooperation and innovation with domestic partners over the past decade. Intel's technical support for customers has always kept pace with the times, and through the continuous upgrading of full-stack solutions, it provides customers with a comprehensive and efficient service system.

Multiple edge application cases were demonstrated on site: from arms, smart classrooms to AI digital humans

The reporter saw the xStream super-resolution transcoder of Videox at the booth. The staff said that this product is based on Intel Xeon Sapphire Rads SP processor and independent graphics card, and uses Intel AI super-resolution software, which can convert 1080HD video into 4K UHD video, realize fine image reconstruction, and image quality close to native 4K, which can help TV stations, operators and other customers quickly build 4K media resource libraries and significantly improve the value of HD content.

Kedong's LLM robotic arm control solution was displayed. This solution is based on LLM reasoning and is deployed on the Core Ultra platform. It also integrates the robotic arm and the servo-driven robotic arm body. The booth staff said that the overall solution is based on the case of Kedong's robot and language-based large models. The case reflects the visual recognition and motion planning capabilities based on AI large models. AI large models accelerate the efficiency of development and meet the needs of lightweight local deployment in industrial scenarios, while ensuring that data does not leave the factory. This technology has been applied in smart factories. The model is about 7GB, and the key requirement is to understand and decompose instructions.

The difficulty of embodied intelligence lies in iteration. The first generation of embodied intelligence needs to be combined with the original robot control, and the underlying layer is still combined with the motion control library; the biggest challenge of the new generation of embodied intelligence, human movements, lies in training, teaching the robot what movements to make.

On-site staff said that PowerLeader AI digital human solution is based on Intel Xeon processors and PowerLeader servers and workstations using Intel CPU as the AI ​​computing power base, which can be widely used in intelligent voice assistants, live broadcasts, online education, retail, etc. In addition, the reporter saw PowerLeader's dual-channel self-developed server PR2715E, which uses Intel's fifth-generation Intel Xeon Scalable Processor, supports 5, PCle5.0 and HBM, and has a performance improvement of up to 50%. It has strong AI computing power and can support 8 single-width or 3 double-width GPU cards.