Home > News > From IDM2.0 to the third-generation Ice Lake scalable processor, Intel’s counterattack is coming

From IDM2.0 to the third-generation Ice Lake scalable processor, Intel’s counterattack is coming

On April 7, Intel released the third-generation Intel® Xeon® Scalable processor (code-named "Ice Lake") in Beijing Shougang Park and announced the launch of a new data center platform based on the processor. Intel's vice president and general manager of China, Wang Rui also talked about the progress of the 7-nanometer process and the "IDM 2.0" that was just put forward in his speech.

At the same time, Intel invited user representatives including Alibaba Cloud, China Mobile, Baidu, Ping An Technology, Tencent Cloud, etc. to share their respective promotion of IT architecture transformation, deployment of cloud-side-end computing solutions, and provision of innovative cloud computing services. Good practice. It has also organized a number of thematic forums including artificial intelligence, cloud computing, big data, 5G cloud network integration, intelligent edge, and high-performance computing.

According to reports, this is Intel's largest foreign activity this year. In the past, the manufacturing process lags behind TSMC and Samsung, followed by the impact of Nvidia and AMD's "back wave". This giant that has been arrogant for decades finally roars and shows the world the giant's counterattack with great momentum.

The performance leap of the third-generation Xeon Scalable processor

According to reports, the third-generation Xeon Scalable processor uses Intel’s 10-nanometer process technology and is the industry’s only data center CPU with built-in artificial intelligence acceleration, supporting end-to-end data science tools and a broad intelligent solution ecosystem. Each third-generation Intel Xeon Scalable processor chip can provide up to 40 cores, and the performance is 2.65 times higher than that of a system that has been deployed for five years. Each slot of the platform can support up to 6TB system memory, 8 DDR4-3200 memory channels and 64 fourth-generation PCIe channels.

Compared with the previous generation, the third-generation Intel Xeon Scalable processor has an average 46% increase in performance on mainstream data center workloads. This product also adds several new and enhanced platform features, including Intel Software Guard Extensions with built-in security features, Intel hardware acceleration for cryptographic operations, and Intel Deep Learning Boost Technology (DL Boost) for artificial intelligence acceleration.

Just less than a month ago, AMD released the third-generation AMD EPYC (Xiaolong) processor (code-named "Milan") series based on the Zen3 architecture online, including 19 models of CPUs with 8 to 64 cores. Using TSMC’s 7-nanometer process for mass production, the third-generation EPYC not only improves IPC performance by about 19%, but also supports PCIe4 and DDR4 memory. At that time, AMD claimed that its EPYC 7763 was 106% ahead of the Intel Xeon Gold 6258R HPC high-performance computing load and cloud computing load. The enterprise load is 117% stronger than Intel's 28-core Intel Xeon Platinum 8280.

At this press conference, Intel is quite tit-for-tat, saying that the performance of the third-generation Xeon in deep learning and inference is 25 times higher than that of AMD EPYC 7763. Among the 20 most common machine and deep learning models identified through investigation, the performance is 1.5 times higher than AMD EPYC. Intel even pulled out the GPU to compare, saying that in the aforementioned survey model, the third-generation Xeon showed 1.3 times the performance advantage of the Nvidia A100 GPU.

Regarding the problem that the number of cores is not as good as competing products, Chen Baoli, vice president of Intel’s marketing group and general manager of data center sales in China, said in communication with the media: “Intel has acceleration instructions for different workloads and supporting products, including artificial intelligence. Some functions such as acceleration, VNNI, security passwords, hardware acceleration, etc. need to be done a lot more when designing the chip. We think that these functions may be able to better meet the needs of customers instead of Just choose an audit."

7nm is expected to tape in the second quarter

At the press conference, Wang Rui also mentioned the IDM2.0 strategy just released on March 23. The growth in demand brought about by the development of the digital economy and the impact of the new crown epidemic and the global trade situation have led to a shortage of global semiconductor supplies in 2021. To this end, Intel released a new IDM2.0 strategy.

The strategy mainly involves three major contents. First of all, Intel’s 7-nanometer research and development is progressing smoothly. It is expected that Tape in’s first 7-nanometer product Meteor Lake is expected in the second quarter of this year. (Tape in refers to the previous step of the final tape out of the chip, usually about a year after entering the tape in stage, the new process will be available.) At the same time, Wang Rui said that combined with advanced 3D packaging technology, Intel will deliver more Customized products to meet the diverse needs of customers.

Secondly, Intel announced that it will further enhance its cooperation with third-party foundries to optimize Intel’s cost, performance and supply, to bring greater flexibility to customers and create unique competitive advantages.

Finally, the strategy announced the reorganization of Intel Factory services to provide foundry services to customers around the world. According to Wang Rui, Intel’s capital expenditure in 2020 reached 14.3 billion U.S. dollars. On this basis, Intel will continue to invest 20 billion U.S. dollars in 2021 to build two new fabs to expand the productivity of high-end processes. In the next stage, production will continue to be expanded in the United States, Europe and other countries to meet the huge global demand for semiconductor chips.

In the past 20 years, Intel has taken the lead in the processor market. Even if it hits a wall in the mobile Internet market and is unable to grasp the baseband, it will not compromise its strong control in the PC and server markets. Especially in the server market, if AMD still has the power to fight Intel in the PC market, then in the server market, it is truly Intel's dominance. However, in the past year, the raging impact of Nvidia, AMD and other "back waves" has made Intel feel an unprecedented "chill."

In 2017, AMD officially released EPYC (Xiaolong) 7000 series server processor products, emphasizing high performance, personalization, and security. In 2019 and 2020, the second-generation AMD EPYC (Xiaolong) series processors of the 7nm process and the "supplements" of the second-generation AMD EPYC (Xiaolong) series processors were successively released. At this point, AMD truly has the "real" strength in Intel's "field" with this giant.

In 2020, AMD launched the brand-new Zen 3 CPU architecture and released the Ryzen 5000 series processors for desktop computers. While still using the 7nm process, the single-core performance surpassed Intel. In the same year, AMD announced its decision to acquire Xilinx through an all-stock agreement for 35 billion U.S. dollars. Its decision to further develop its ambitions in the data center was revealed, and the data center is Intel's core battlefield.

At the same time, following the successful acquisition of Mellanox by defeating Intel the previous year, Nvidia announced the acquisition of Arm in 2020, which is even more powerful in the development of its data center business. The market also gave Nvidia unprecedented expectations. Nvidia's market value surpassed Intel in July of that year.

All this is closely related to Intel's backwardness in process technology in the past few years. With the launch of Intel’s IDM 2.0 strategy and the release of a highly competitive new data center platform, it has officially stepped out to fight back AMD and Nvidia with great fanfare. However, whether these measures are sufficient to allow Intel to regain its glory remains to be verified by time.