Intel's Song Jiqiang: Collaborative innovation between cloud, edge and end to accelerate the application of intelligent computing

On November 18, Song Jiqiang, vice president of Intel Research and director of Intel China Research Institute, delivered a speech titled "Product Design and Process Technology Innovation for the 'Intelligent Computing Era'" at the 21st China International Semiconductor Expo (IC China 2024). He shared his insights on the development trends of intelligent computing technology and introduced how Intel can accelerate the implementation of intelligent computing from cloud to end through product and technology innovation to promote the development of the digital economy and industrial transformation and upgrading.

Song Jiqiang, Vice President of Intel Research and Director of Intel China Research Institute, delivered a keynote speech at IC China 2024

According to relevant technology development research reports, the number of parameters in large language models is showing a rapid growth trend. Models of this size will bring excellent performance, and it requires collaboration from all parties to implement them in a sustainable manner in application scenarios in thousands of industries. Intel CEO Pat Gelsinger pointed out that every company will become an AI company, every device will become an AI device, and everyone can benefit from the magical power of AI technology.

Song Jiqiang said that the implementation of intelligent computing requires not only strong semiconductor computing power support, but also comprehensive technological innovation from cloud to end. Different levels of computing resources have different uses in AI applications. From clients to edges to data centers, as the complexity and scale of tasks increase, the required computing resources are also increasing. Different types of workloads can achieve "twice the result with half the effort" only when they run on more suitable hardware, saving costs and improving efficiency.

For example, at the client level, AI PCs are suitable for performing lightweight inference tasks; at the edge, workstations are suitable for model fine-tuning and inference tasks, and service clusters are suitable for lightweight training and peak inference tasks; larger-scale training and inference loads need to be completed on super service clusters and giant service clusters in the data center.

From the most basic process and packaging technology to products for clients and servers, to the entire ecosystem, Intel is promoting the further implementation and development of intelligent computing through comprehensive product and technology innovation.

Process and packaging technology innovation

Song Jiqiang introduced that in terms of underlying technology, Intel will achieve the "Five process nodes in four years" plan in 2025 and regain process leadership through Intel 18A. Currently, the first batch of products based on Intel 18A, the client processor Panther Lake and the server processor Clearwater Forest, have been shipped, powered on and successfully started the operating system, and are expected to start mass production in 2025.

Intel 18A wafer

In Intel 18A, Intel has successfully integrated two innovative technologies. One is the RibbonFET all-around gate transistor, which can provide faster transistor switching speed with the same drive current in a smaller footprint. The other is the PowerVia back-side power supply technology, which moves the power line to the back of the transistor and separates it from the signal line, which helps to increase transistor density and improve power supply. Intel 18A will also be open to Intel foundry customers.

Combining Intel's advanced packaging capabilities, 2.5D technology EMIB (Embedded Multi-Chip Interconnect Bridge) and 3D technology Foveros, as well as future Foveros Direct (hybrid bonding solution), glass substrates and optoelectronic co-sealing, these underlying technology innovations will help Intel and its foundry customers create chips with higher performance and lower power consumption, and flexibly meet the computing power requirements of intelligent computing.

Products and Ecosystem

Song Jiqiang also shared Intel's product progress in 2024. Diversified chips can meet the differentiated needs of different users and customers and accelerate innovation in intelligent computing. The modular SoC architecture design brings high flexibility and scalability to these products. In addition to products, Intel is also actively working with ecosystem partners such as original equipment manufacturers (OEMs) to further promote the implementation of intelligent computing through a full-stack, open software ecosystem.

In terms of server processors, Intel has launched the Intel Xeon 6 Energy Efficient Core processor code-named Sierra Forest, and the Intel Xeon 6 Performance Core processor code-named Granite Rapids. The former is specifically designed for high-performance optimization required for high core density and scale-up tasks, such as cloud-native applications, content distribution networks, etc., which can significantly reduce energy costs while providing efficient computing power; the latter is designed for computing-intensive workloads such as AI, data analysis, and scientific computing. The performance has been greatly improved compared to the previous generation of products, and with more cores, double memory bandwidth, and built-in AI acceleration functions, it meets various intelligent computing challenges from the edge to the data center to the cloud environment. At the same time, the Xeon 6 product architecture is compatible, sharing software stacks and an open ecosystem of software and hardware suppliers.

Intel Xeon 6-core processor

In terms of client processors, Intel released the Core Ultra 200V series processors codenamed Lunar Lake and the Core Ultra 200S processors codenamed Arrow Lake to accelerate AI PC innovation on mobile and desktop ends respectively . Currently, Intel has shipped more than 20 million AI PC devices. In terms of the ecosystem, it has supported more than 100 ISVs, more than 300 AI applications and more than 500 AI models. IDC predicts that AI PCs will account for half of the market in 2025, and this proportion will reach 100% in 2030.

Core Ultra 200V series processors

In particular, in response to Chinese customers' demand for energy-efficient, customized packaging solutions, Intel has announced the expansion of its Chengdu packaging and testing base, adding packaging and testing services for server chips, and setting up a customer solution center as a one-stop platform to promote enterprise digital transformation and accelerate the implementation of industry applications. Song Jiqiang said that Intel will increase its support for Chinese customers, improve response speed, and provide customized solutions based on Intel architecture and products.