Intel held a product launch event in China on the 15th, announcing the launch of the fifth generation Xeon processor and new Core Ultra processor for artificial intelligence. These two products, along with Intel's latest artificial intelligence (AI) chip Gaudi 3, were also released in the US market at the same time.
During the press conference, domestic technology giants such as JD Cloud, Alibaba Cloud, Ping An Technology, Volcano Engine, Baichuan Intelligence, Lenovo, and Baidu all provided platforms for Intel products. "More than 20 hardware partners and 7 cloud service providers in China have prepared new servers and cloud services based on the fifth generation Xeon processors," said Chen Baoli, Vice President and General Manager of Intel Data Center and Artificial Intelligence Group in China, at a press conference on the 15th.
Although Intel did not officially launch the Gaudi3 at the press conference in China, the product still caught the eye at the US launch event. Intel expects the Gaudi3 to be officially launched in 2024, and the company claims that the performance of the Gaudi3 will be better than Nvidia's H100, but this claim has not been independently verified. In addition to Nvidia's H100, the Gaudi 3 also faces competition from AMD's "state-of-the-art AI accelerator" MI300, which was also delivered to customers in 2024.
Official information shows that Intel Gaudi3 will adopt a 5nm process, with a bandwidth 1.5 times that of the previous generation Gaudi2 (7nm process), BF16 power 4 times, and network computing power 2 times. Gaudi3 is expected to be equipped with up to 128GB of HBM3e memory, significantly improving AI learning and training performance.
In the current AI arms race, Nvidia is currently the leader in market share, and on the 14th of last month, the company officially released the new generation of artificial intelligence computing chip H200. Official information shows that the H200 chip is built on the NVIDIA Hopper architecture, with a processing speed of 4.8TB/s and 141GB of memory. Compared to the previous generation product A100, the capacity of H200 has almost doubled. Nvidia states that in order to apply generative artificial intelligence, it is necessary to use large capacity GPUs to efficiently process large amounts of data.
Focusing on NVIDIA, AMD unveiled the MI300 series of products in a high-profile manner earlier this month. It is reported that the MI300X chip has over 150 billion transistors, with a memory density 2.4 times that of the current market leader Nvidia H100 and a memory bandwidth 1.6 times that. AMD CEO Su Zifeng claimed that this new chip is comparable to Nvidia's H100 in terms of training artificial intelligence software and performs better in reasoning, but this claim has not been independently verified.
In 2024, the AI chip market will face competition from three giants: NVIDIA H200, AMD MI300 series, and Intel Gaudi3. Nvidia's dominance in the field of AI chips will face challenges from two other manufacturers.
Compared to Nvidia, Intel, as a latecomer, hopes to overcome Nvidia's pain points through its own improved downstream product ecosystem. Therefore, at the press conference in China, Intel released the fifth generation Xeon processor for enterprises, as well as the Core Ultra processor for personal computers (PCs).
Intel introduced that the fifth generation Xeon processor can bring better AI performance for data center, cloud, network and edge computing, with an average increase of 21% over the previous generation of general computing performance, and an average increase of 36% in energy consumption per watt. The fifth generation Xeon processors have improved inference and fine-tuning capabilities by up to 42% on models with 20 billion parameters.
On the other hand, Intel hopes to promote the prosperity of its upstream hardware ecosystem by vigorously promoting the implementation of AI on personal computer terminals. In September of this year, Intel CEO Pat Kissinger first proposed the concept of AI PC. At that time, Kissinger pointed out that AI would work closely with PCs through the cloud, fundamentally changing, reshaping, and reconstructing PC experiences, unleashing people's productivity and creativity. He predicts that the shipment scale of AI PCs will reach 100 million units within the next two years.
Intel has also released a new Core Ultra processor for personal devices. It is reported that this processor is suitable for laptops and desktops, allowing the aforementioned terminals to directly apply AI functions. During the press conference, Intel emphasized that it is collaborating with over 100 software vendors to bring hundreds of AI enhanced applications to the PC market. These highly creative, productive, and fun applications will change the user experience of PCs.
After taking over Intel, Kissinger has been hoping to revive the company's declining business during the downturn in the personal computer market through the AI boom. However, Intel's artificial intelligence business remained under pressure in the first three quarters of this year. According to the company's financial report, in the third quarter of this year, the company's data center and AI (artificial intelligence) revenue was $3.814 billion, a decrease of 10% year-on-year. The department's revenue in the second quarter was $4 billion, a year-on-year decrease of 15%, and a year-on-year decrease of 39% in the first quarter.
During the press conference, Intel stated that due to its continuously growing and validated performance advantages, as well as highly competitive Total Cost of Ownership (TCO) and pricing advantages, Gaudi3 is expected to occupy a larger market share in 2024 with its AI chip suite.
According to the White Paper on the AI PC Industry (China) released by research firm IDC and Lenovo, although consumer electronics have slowed down, the report believes that the AI PC market is expected to see new growth. In the next 5 years, the proportion of new machines assembled by AI PCs in the Chinese PC market is expected to increase from 8.1% in 2023 to 85% by 2027, becoming the mainstream of the PC market. This means that the installed capacity of AI PCs will increase by more than 10 times within 5 years.