The first shot in the CPU market in 2023 -- Intel's fourth generation Xeon scalable processor release

With the end of CES, the semiconductor market has begun.  On January 11, 2023 (Beijing time), Intel officially launched the fourth generation Intel Xeon Expandable processor and Intel Xeon CPU Max family products, as well as Intel data center GPU Max family products for scientific computing and artificial intelligence.   Is officially started in 2023, the new product release of the semiconductor market, as the absolute king of the processor field Intel, its Xeon series is its signature in the field of data center.   First released in 1998, the Xeon series was created for the server CPU space, and for nearly 25 years, the Xeon series has led the development of the server CPU market.

For many consumers, server CPU may not be with your life, but, in the view of senior PC installed consumers, Xicheng series is famous "foreign garbage", some old Xicheng CPU, after retiring in the server, can still be bought with affordable prices, continue to shine in the home machine, We can see the stability and strong performance of the Xeon series.  Let's take a look at the new Xeon series.  If you want to assemble computers at a low cost in five years, you will have to use the fourth generation Xeon bought from overseas.The codename for Xeon is Sapphire Rapids.  intel's decision to name the product by Rapids, rather than Lake, is supposed to indicate that Intel wants to distinguish the Xeon series from the average consumer CPU.  In the future, the Lake and Rapids will be available as Intel's commercial and consumer product lines, providing Intel with a strong presence in the CPU market.

First, and most notably, the SPR is Intel's first Xeon processor designed based on Chiplet.  In terms of packaging, Intel Xeon 4 combines up to four Intel 7 process tiles in a single package.  The use of Intel EMIB (Embedded multi-chip Interconnect and bridge) packaging technology for connectivity, so many new technology applications, fully reflects Intel's ambitions in this generation Xeon.  Xeon products include 52 cpus that support up to 60 cores, PCIe 5.0, DDR5 memory, and CXL 1.1 ports (type 1 and 2 devices).  It provides up to 80 PCIe 5.0 channels, up to 1.5TB DDR5-4800 memory, and up to 350W TDP.

In terms of CPU accelerators, with the adoption of new technology integration accelerator (DL Boost, AMX, QAT, DSA, DLB for artificial intelligence acceleration).

The fourth generation Intel Xeon expandable processor optimizes the AI work. Compared with the previous generation, the AI performance will be significantly improved. Thanks to Intel Advanced Matrix extension AMX, the fourth generation Intel Xeon expandable processor improves the PyTorch real-time reasoning and training performance by 10 times. Intel has also extended and enhanced natural language processing in particular, increasing the speed of large language models by up to 20 times. In addition, this generation will effectively improve the power consumption ratio compared to the previous generation. According to Intel officials, the average performance per watt efficiency of the target workload can be improved by 2.9 times. In the optimized power consumption mode, each CPU can save up to 70W power consumption, while minimizing the performance loss of specific workloads. Total cost of ownership (TCO) can be reduced by 52% to 66%. With so many accelerators in the mix, Intel seems to want to move beyond traditional CPU products that rely on stacking cores and pushing up CPU frequencies, and move its server cpus to superior efficiency for specific tasks.

When it comes to the data center, Xeon's fourth generation CPU Max series integrates 47 small chips and more than 100 billion transistors in a single product, delivering higher throughput for challenging workloads such as physics, financial services and life sciences. The LAMMPS processing performance can be improved by 12.8 times in life and materials science. In terms of data center security, Intel claims that this Xeon will provide users with the industry's most comprehensive portfolio of confidential computing products, which will provide today's minimal attack coverage for confidential computing in private, public and cloud-to-edge environments. Not only that, but Intel's new virtual machine isolation technology, Intel Trust Domain Extension (Intel TDX), will make its debut in Microsoft Azure, Alibaba Cloud, Google Cloud, and IBM Cloud, enabling existing applications to be ported to confidential environments.

Intel's fourth-generation Xeon will sell for $415 (about 2,814 yuan), with a top price of $17,000. Intel said, Amazon AWS, Cisco, Cloudera, CoreWeave, Dell Technologies, Dropbox, Ericsson, Fujitsu, Google Cloud, HP, IBM Cloud, Inspur Information, IONOS, Lenovo, Los Alamos National Laboratory, Microsoft Azure, Nvidia, Oracle Customers and partners such as Cloud, OVHcloud, phoenixNAP, RedHat, SAP, AMD, Telefonica and VMware will adopt Xeon 4 processors.

In the end, Mr. Chen Baoli, Vice President and general Manager of China, Intel Data Center and AI Group, said, "Intel will invest a lot of software services and customer cooperation, so that more industrial partners in China, whether it is Internet manufacturers, ISV partners, or software partners, can better use the fourth generation Xeon processor functions."