On September 19, local time, the 2023 Intel on Technology Innovation Conference opened in San Jose, California, USA. At this conference for developers, Intel announced a series of new technologies aimed at making AI ubiquitous and making it more pervasive in all workloads from the client and edge to the network and cloud. application.
Pat Gelsinger, CEO of Intel Corporation, delivered the opening keynote speech. He said: "AI represents the arrival of a new era. AI is giving rise to a new era of global growth. In this new era, computing power is rising. play a more important role in a better future for everyone. For developers, this will bring huge social and business opportunities to create more possibilities and create solutions for the world’s major challenges. And to the benefit of every human being on earth.”
In the opening keynote speech, Gelsinger demonstrated how Intel is adding AI capabilities to its various hardware products and promoting the popularization of AI applications through open, multi-architecture software solutions. Kissinger also emphasized the role of AI in promoting the "core economy", which refers to "the growing economic form driven by chips and software." Today, chips form a $574 billion industry and drive the global tech economy of approximately $8 trillion.
Latest Advances in Processing, Packaging and Multi-die Solutions
The vigorous development of the "core economy" begins with the innovation of chip technology. Kissinger said that Intel's "Five Process Nodes in Four Years" plan is progressing smoothly. Intel 7 has achieved large-scale mass production, Intel 4 is ready for production, and Intel 3 is also progressing as planned, with the target being the end of 2023.
During the keynote speech, Gelsinger also demonstrated the first batch of test chips for the Intel Arrow Lake processor based on the Intel 20A process node. Arrow Lake will be launched for the client market in 2024. Intel 20A will be the first process node to apply PowerVia backside power supply technology and the new all-around gate transistor RibbonFET. The Intel 18A process node, which will also use these two technologies, is also progressing as planned and will be ready for production in the second half of 2024.
In addition to manufacturing processes, another way for Intel to advance Moore's Law is to use new materials and new packaging technologies, such as glass substrates. This is a breakthrough that Intel just announced this week. Glass substrates will be introduced in the late 2020s, continuing to increase the number of transistors in a single package, helping meet the needs of data-intensive, high-performance workloads such as AI, and continuing to advance Moore's Law after 2030.
Glass substrate test unit
Intel also demonstrated test chip packaging based on the Universal Core Interconnect Express (UCIe) open specification. Kissinger said that the next wave of Moore's Law will be driven by multi-chip packaging technology, and if open standards can solve the obstacles to IP integration, it will soon become a reality. The UCIe standard, launched last year, will allow chips from different manufacturers to work together to meet the expansion needs of different AI workloads with new chip designs. Currently, the UCIe open standard has been supported by more than 120 companies.
The test chip integrates Intel UCIe IP chips based on the Intel 3 process node and Synopsys UCIe IP chips based on the TSMC N3E process node. These die are interconnected through EMIB (Embedded Multi-die Interconnect Bridge) advanced packaging technology. Intel Foundry Services, TSMC and Synopsys have joined forces to promote the development of UCIe, reflecting the commitment of the three parties to support a core ecosystem based on open standards.
Powering AI everywhere with performance improvements
Gelsinger highlighted the variety of AI technologies currently available to developers on Intel platforms and how related technologies will be significantly expanded in the coming year.
The recently announced MLPerf AI inference performance test results further reinforce Intel’s commitment to covering AI models of all sizes, including larger and more challenging generative AI and large language models. The test results also prove that Intel® Gaudi®2 accelerator can provide an excellent solution to meet AI computing needs. Kissinger also announced that a large-scale AI supercomputer will be built entirely using Intel Xeon processors and 4,000 Intel Gaudi2 accelerators, and Stability AI is its main customer.
Alibaba Cloud CTO Zhou Jingren explained how Alibaba uses fourth-generation Intel® Xeon® Scalable processors with built-in AI accelerators for its generative AI and large language models, the "Alibaba Cloud Tongyi Qianwen Large Model." Zhou Jingren said that Intel technology "significantly shortens the model response time, with an average acceleration of up to 3 times."
Intel also previewed the next-generation Intel Xeon processors and revealed that the fifth-generation Intel® Xeon® processors will be released on December 14, which will improve performance and storage speeds for global data centers at the same power consumption. . In addition, the energy-efficient E-core processor Sierra Forest will be available in the first half of 2024. With 288 cores, the processor is expected to increase rack density by 2.5 times and improve performance per watt by 2.4 times compared to the fourth generation Xeon. Following the release of Sierra Forest is Granite Rapids, a high-performance performance core (P-core) processor. Compared with the fourth-generation Xeon, its AI performance is expected to increase by 2 to 3 times2.
5th Generation Intel Xeon Processor
Looking forward to 2025, the next-generation Xeon energy-efficient core processor, codenamed Clearwater Forest, will be manufactured based on the Intel 18A process node.
Launched AI PC equipped with Intel Core Ultra processor
AI will also become more personal. Kissinger said: "AI will fundamentally change, reshape and reconstruct the PC experience through the close collaboration of cloud and PC, releasing people's productivity and creativity. We are moving towards a new era of AI PC."
This new PC experience will soon be demonstrated on the upcoming Intel Core Ultra processor, code-named Meteor Lake. The processor features Intel's first integrated neural network processor (NPU) to bring energy-efficient AI acceleration and local inference experiences to PCs. Gelsinger confirmed that Core Ultra will also be released on December 14.
The Core Ultra processor is a turning point in Intel's client processor roadmap: it is the first client die design to use Foveros packaging technology. In addition to the significant improvements in performance-power consumption ratio of NPU and Intel 4 process nodes, this processor also brings independent graphics card-level performance by integrating Intel Ruixuan™ graphics.
Intel Core Ultra Processor
On the stage, Gelsinger demonstrated many usage scenarios of the new AI PC, and Acer Chief Operating Officer Gao Shuguo introduced Acer laptops equipped with Core Ultra processors. Gao Shuguo said: "We worked with the Intel team to jointly develop a set of Acer AI libraries through the OpenVINO toolkit to make full use of the Intel Core Ultra platform. We also jointly developed the AI library and finally brought this product to users."
Let developers become drivers of the core economy
Pat Kissinger said: Future artificial intelligence must contribute to the improvement of accessibility, scalability, visibility, transparency and trust of the entire ecosystem.
To help developers create this future, Intel announced:
● Intel Developer Cloud Platform is fully online: Intel Developer Cloud Platform helps developers take advantage of the latest Intel software and hardware innovations for AI development (including the Intel Gaudi2 accelerator for deep learning), and authorizes them to use Intel’s latest hardware platforms, Such as fifth generation Intel® Xeon® Scalable processors and Intel® Data Center GPU Max Series 1100 and 1550. When using the Intel Developer Cloud Platform, developers can build, test, and optimize AI and scientific computing applications. They can also run AI training, model optimization, and inference workloads from small to large to achieve high performance and high efficiency. The Intel Developer Cloud Platform is built on oneAPI, an open programming model that supports multi-architecture and multi-vendor hardware. It provides developers with hardware choices and gets rid of proprietary programming models to support accelerated computing, code reuse and Meet portability requirements.
● Intel releases OpenVINO tool suite version 2023.1 released: OpenVINO is Intel’s AI inference and deployment runtime tool suite, providing developers with high-quality choices on client and edge platforms. This release includes pre-trained models optimized for integration across operating systems and a variety of different cloud solutions, including multiple generative AI models such as Meta’s Llama 2 model. Companies including ai.io and Fit:Match demonstrated on-site how they use OpenVINO to accelerate applications: ai.io uses OpenVINO to evaluate athlete performance; Fit:Match uses OpenVINO to revolutionize the retail and health industries to help consumers find Clothes that fit better.
● Strata project and development of edge-native software platform: The platform will be launched in 2024 and provide modular components, high-quality services and product support. This is a way to scale out the infrastructure required for intelligent edge and hybrid AI and integrate vertical applications from Intel and third parties within a single ecosystem. The solution will enable developers to build, deploy, run, manage, connect and secure distributed edge infrastructure and applications.
Through a series of new technologies and products, Intel unveiled the 2023 Intel on Technology Innovation Conference. Next, Intel Chief Technology Officer Greg Lavender will deliver a keynote speech at 9:30 a.m. Pacific time on September 20 (0:30 a.m. on September 21, Beijing time) to introduce how Intel can provide developers with more ways to use AI. Create opportunities in the field and accelerate the integration of AI and security.