On September 20, Beijing time, Intel hosted the "ON Technology Innovation Summit," where they showcased a range of products, including the Core Ultra processor designed for AI PCs, the AI computing chips Gaudi2 and Gaudi3, the fifth-generation Xeon processors, advanced manufacturing processes, and the latest in packaging technology.
In the opening remarks, Intel's CEO, Pat Gelsinger, discussed the scale of the semiconductor industry and the tech economy, emphasizing the value of computational power within them.
According to Intel's data, the semiconductor industry has reached a market size of $574 billion, while the global tech economy based on chips has soared to $8 trillion, introducing the concept of a "Silicon Economy."
Gelsinger believes that AI signifies the dawn of a new era, bringing immense opportunities. More abundant, powerful, and cost-effective processing capabilities are key components of future economic growth. He mentioned that the number of connected devices is expected to quadruple in the next five years and increase fifteen-fold in the next decade.
Core Ultra Set to Launch on December 14
Earlier in June, Intel announced a brand upgrade strategy. The "Core Ultra" is positioned as a flagship, subdivided into Core Ultra 9, Core Ultra 7, and Core Ultra 5. The traditional "Core" is positioned as mainstream, categorized as Core 7, Core 5, and Core 3.
At the "Intel ON Technology Innovation Summit," the Core Ultra processor, codenamed Meteor Lake, was unveiled. It uses Intel's 4th generation process and 3D Foveros packaging technology, integrates Intel's Iris Xe graphics, and is set to launch on December 14. Acer's Swift laptop will be the first to feature this new processor, equipped with Acer's AI software library developed based on the OpenVINO toolkit.
While detailed specifications for the Core Ultra processor were not disclosed at the summit, previous information suggests that this processor will be crafted using both Intel's 4th generation process and TSMC's 5nm process. It will utilize the second-generation hybrid architecture technology, with the P-core adopting the new Redwood Cove architecture and the E-core using the Crestmont architecture.
Additionally, the Core Ultra 7 155H processor, based on leaked benchmark data, will have 6 performance cores and 10 efficiency cores, totaling 16 cores and 22 threads. It will have a base frequency of 3.8GHz, a maximum turbo frequency of 4.8GHz, 4MB of L2 cache, and 24MB of L3 cache.
Intel also demonstrated the AI processing capabilities of an AI PC based on the Luna Lake processor at the summit. This involved generating a song titled "AI Taylor Swift" from text in a matter of seconds while offline. Luna Lake is expected to launch in 2024, utilizing Intel's 18A process and the same 3D Foveros packaging technology as the Core Ultra processor.
The fifth-generation Intel Xeon processor is set to launch on December 14. The energy-efficient E-core processor, Sierra Forest, which boasts higher energy efficiency, is slated for a release in the first half of 2024. Compared to its fourth-generation counterpart, the 288-core high-efficiency processor is expected to increase rack density by 2.5 times and improve performance per watt by 2.4 times. The high-performance P-core processor, Granite Rapids, will be released after Sierra Forest and is anticipated to enhance AI performance by 2 to 3 times compared to the fourth-generation Xeon.
Intel also mentioned that the next-generation Xeon efficiency core processor, codenamed Clearwater Forest, will be manufactured based on the Intel 18A process node.
Gaudi2 Partners with Stability AI
In May, Intel introduced the Gaudi2 AI computing chip based on the 7nm process and subsequently launched it for the Chinese market in July. At the summit, Gelsinger shared updates on the processor's application progress. It was revealed that Stability AI, using Intel's Xeon processors and 4,000 Intel Gaudi2 AI computing chips, has built a large-scale AI supercomputer, ranking among the top 15 globally in terms of performance.
Gaudi2 features 24 programmable Tensor Processor Cores (TPCs), 96GB HBM2e memory, and 24 100GbE ports. Official data indicates that its performance in video and natural language processing model tests, such as RestNet50 Training Throughput and BERT Training Throughput, is approximately twice that of NVIDIA's A100.
Beyond reviewing Gaudi2's progress, Gelsinger mentioned that the next-generation AI computing product, Gaudi3, will be based on a more advanced 5nm process. Compared to its predecessor, Gaudi3's BF16 performance will increase fourfold, computational performance will double, network bandwidth will improve by 1.5 times, and HBM capacity will also increase by 1.5 times.
Aiming for "World's Second" in Chip Manufacturing by 2025
In 2022, the U.S. introduced the "Chip Act," vigorously promoting the return of advanced process chip manufacturing. Before this, Intel had already begun its preparations.
In 2021, Intel announced its IDM 2.0 strategy, pushing forward with its foundry business. They set a goal to mass-produce five process nodes within four years and aim to become the world's second-largest wafer foundry by 2025. Gelsinger reviewed the progress of the foundry business, and related goals are being achieved step by step.
So far, Intel 7 has been mass-produced, the Core Ultra based on Intel 4 is about to launch, and Intel 3 is progressing as planned. The Angstrom-level process nodes, Intel 20A and Intel 18A, are set for mass production in the first and second halves of 2024, respectively.
Intel stated that Intel 20A will be the first process node to apply PowerVia backside power supply technology and the new RibbonFET all-around gate transistor.
Earlier in March, Intel announced that Intel 20A and Intel 18A had successfully taped out, meaning specifications, materials, and performance targets had been largely achieved.
Furthermore, Intel's foundry service has 43 potential partners testing chips, with at least seven from the global top 10 chip customers, including ARM and Ericsson.
Regarding advanced packaging technology, besides supporting the coexistence of different processes and structures with Foveros 3D packaging technology, Intel also revealed the development progress of semiconductor glass substrate packaging material technology.
Officially, compared to the current industry-standard organic substrates, glass offers unique performance advantages in terms of flatness, thermal stability, and mechanical stability. This will allow chip architects to package more chiplets in a single package, achieving improvements in performance, density, flexibility, and reductions in cost and power consumption.
Intel claims that its glass substrate technology can increase the chip area in a single package by 50%, allowing for the inclusion of more chiplets. It's understood that the glass substrate can also be used alongside traditional organic substrates to enhance structural integrity.
