Processor architecture evolves with data, and we are currently in the midst of a data explosion led by four superpowers – ubiquitous computing, ubiquitous connectivity, artificial intelligence, and infrastructure from the cloud to the edge In this era, various types of data are generated every second, such as scalar, tensor, vector, spatial data, etc. To cope with these explosive growth of big data, what kind of processors will data centers need in the future?
On October 21st, at the 2021 CCF National High Performance Computing Academic Conference held in Zhuhai, Liang Yali, Vice President of Intel Corporation Marketing Group and General Manager of China Industry Solutions Department, shared the theme of “Accelerating Technology Leadership with High Performance Computing” Report. We can get a glimpse of the future processor architecture of the E-class data era from her sharing and other speaker reports.
Liang Yali, Vice President of Intel Corporation Marketing Group and General Manager of China Industry Solutions Department
Liang Yali said: “Today, the intelligence of everything is getting closer and closer to a reality. This is actually a huge change, but it has also led to an exponential increase in people’s computing and demand. Technology is becoming more and more important to all aspects of human life, and high Exascale computing power in the performance computing category is one of the most important infrastructures to support this change.”
She pointed out that Intel is not only a large-scale manufacturing company with both depth and breadth in software, chip and platform, packaging and process technology, but also a leading developer of semiconductor process technology. It has unique advantages in innovation and can be summarized as four “super technology forces” driving digital transformation:
The first is ubiquitous computing. In all fields of the world, computing has become a pervasive key factor that penetrates into all kinds of devices and all kinds of fields.
Second, ubiquitous connectivity. Whether it is a mobile phone, in fact, from Bluetooth to short-distance communication, to long-distance narrowband, 5G broadband, wired transmission, and the process of low-latency networking that is currently being developed and considered, various cities are all in order to make The realization that all devices can be ubiquitous, the connection between people, the connection between people and devices, and the connection between devices and devices has actually been slowly realized, which is the so-called connection upgrade.
Third, infrastructure from cloud to edge. Being able to build ubiquitous computing connections redefines a topology that will define the entire network.
Fourth, artificial intelligence. Among the extended network devices and structures, intelligence is the most prominent requirement. Although artificial intelligence is only one of the computing workloads, Intel proposed it separately because Intel believes that artificial intelligence will redefine the definition. The entire computing architecture and the entire computing process.
At the same time, each superpower will have its own characteristics and will complement each other to create new possibilities. Therefore, combining these four points, we can foresee that we will think about the bottom layer, technology, architecture, and evolution in the process of continuous evolution of the entire digital economy in the future. From these four points, we can also see that in the future, in the digital economy, it will definitely follow four laws, the first connection upgrade, the second thinking upgrade, the third experience upgrade, and the fourth security upgrade.
At the event, Intel invited Trish Damkroger, vice president of Intel’s accelerated computing system and graphics card group and general manager of the high-performance computing business unit, to share Intel’s response to the high-performance computing product line now and in the future by playing a video. these unprecedented challenges.
Trish Damkroger, Vice President, Accelerated Computing Systems and Graphics Group and General Manager, High Performance Computing Group, Intel Corporation
“Intel has been adding HPC-specific capabilities to its Xeon® processor roadmap for the past few years,” she said. “But for Intel to become a HPC leader, a shift to HPC-focused architectures is required. and system.”
The latest 3rd generation Intel released in April this year? Xeon? Scalable processors provide users with reliable performance and a high degree of flexibility, and bring a huge leap in performance to people’s high-performance computing systems. This is thanks to improvements in the processor’s core architecture, with a 20 percent increase in the number of instructions per clock, in addition to increased cache size. Intel also offers up to 40 cores and 8 memory channels, 3rd Gen Xeon? Scalable processors provide up to 6TB of system memory capacity. Xeon? The processor remains the only x86 processor in the data center with built-in AI acceleration, with built-in flexibility features. For example unique Intel? Speed Select technology, which basically combines three CPUs into one, so people can use a single processor to meet more diverse needs.
Trish Damkroger also introduced us to Intel’s next-generation processor architecture, Sapphire Rapids, which will provide outstanding out-of-the-box performance and functionality for a variety of workloads, usage models, and deployment models in the data center. On the base server node, Sapphire Rapids introduces a new performance core microarchitecture with significantly improved IPC for excellent out-of-the-box performance and a number of new architectural features that dramatically improve data-parallel performance.
At the heart of Sapphire Rapids is a new modular SoC architecture that establishes next-generation scalability while maintaining the strong attributes of a balanced SoC that, despite being composed of multiple chip Modules, remains logically monolithic. Sapphire Rapids is the first Xeon product built using Silicon Bridge technology, which enables Intel to build highly scalable and well-balanced processors that have not been possible before.
Liang Yali said at the same time, “Technology has been integrated into everyone’s life, so different application scenarios correspond to different applications, different applications correspond to different workloads, and different workloads must have the most efficient computing system, and the future must be corresponding to XPU. If you need general-purpose computing, you can use CPU. If there is an acceleration scenario, you can use FPGA, ASIC chip, or the PAC we will release in the future. Different application loads will definitely have the most suitable and efficient system for it.”
This also proves why OneAPI is so important? “OneAPI is very important because we don’t want each system to correspond to different libraries, compilers and architectures. We want to have a framework that solves these problems and allows us to focus on the problems that should be focused most.”
In recent years, as high-performance computing has gradually become more and more industrial applications, its popularization and pervasiveness have also attracted more and more attention in the industry. For example, in the medical field, Intel is working with some industry partners to do cutting-edge “no-man’s-land” exploration. For this kind of “cross-border” attempt, Lu Gang, Center for Translational Medicine, Ruijin Hospital, introduced that on the one hand, we need talents with relatively specialized technical knowledge and background in bioinformatics, and at the same time, we need to have a very good understanding of HPC system architecture and technology. Clearly, such talent is very scarce. We and Intel have carried out in-depth cooperation in the design and construction of the entire platform. Intel’s technical strength and the professional capabilities of our experts in the field of bioinformatics can be well integrated, making our platform the first platform in the field of bioinformatics that can make breakthroughs in storage.
Looking ahead to the future of HPC development, Intel will not only provide a comprehensive product portfolio with capabilities ranging from general-purpose computing to specialized acceleration, to unique persistent memory and exascale computing distributed storage, high-performance interconnects, and innovative security. It will also work together with more ecological partners to make technologies work together seamlessly, thereby accelerating the popularization and universalization of HPC, and promoting HPC to bloom in more industries.