Comprehensive computing ambition! An article explains the new Armv9 architecture that took ten years to develop

Recently, at the 9th Annual China Electronic ICT Media Forum and 2021 Industry and Technology Outlook Seminar hosted by Evison, Zou Wei, senior FAE manager of Arm Technology, gave an in-depth interpretation to the industry of Arm's new generation Armv9 architecture, which was newly released after ten years of polishing to target different levels of computing power requirements and machine learning (ML) development paths. It is not only another milestone in the evolution of Arm architecture, but will also become the foundation for Arm to promote industry innovation in the next ten years and beyond.

Zou Wei, senior FAE manager of ARM Technology, gave a speech

Armv9: How it will change the way high-performance computing is done?

Looking at the development of semiconductor design over the past decade, especially in the field of terminal consumption, the productivity improvement brought by computing power has put a lot of computing power in the hands of every consumer through the mobile Internet. At the same time, the traditional process of relying on manual data collection and analysis is evolving into machine-to-machine data collection, storage, analysis and services in the fifth wave of computing. The resulting massive data requires more powerful computing power in the cloud and edge, and the acquired data needs to be analyzed, constructed and decided through the data center.

In addition to becoming the mainstream architecture of the mobile wave, Arm has entered new scenarios such as the edge and cloud. Its unique positioning can provide a solid security foundation for the industry while ensuring that developers use Arm's security technology in the simplest and fastest way. According to Zou Wei, as of the end of 2020, a total of 180 billion chips based on Arm IP have been sold, covering every corner and every place in all walks of life. The shared data must have one or more links calculated by the Arm processor, whether it is an IoT device for data collection, or a daily mobile phone, smart computer, or even a server, supercomputer, etc.

According to Arm's prediction, partners will ship more than 300 billion Arm-based chips in the next decade, and 100% of shared data will be processed on Arm chips, whether on the terminal or in the cloud. Therefore, a new processor architecture is urgently needed to change the way computing is done in the next decade to provide higher performance, improve security features, and enhance workload capabilities.

Zou Wei introduced the new Armv9 architecture features on site

"To this end, Arm has officially launched the next-generation architecture Armv9 for the next decade. It is based on the widely used Armv8 and continues to use AArch64 as the benchmark instruction set, maintaining backward compatibility. On this basis, it has made improvements in security, AI and ML, scalable vector extensions and DSP, expanding the scope of application." Zou Wei said, "Processors implemented by the Armv9 architecture can be used in mobile computing, HPC high-performance computing, automotive and AI markets to meet the world's demand for increasingly powerful security, artificial intelligence and dedicated processing. This means that computing technology based on the Arm architecture will also gain a leading position in markets other than smartphones, and with the huge scale advantages brought by the mobile ecosystem, it will create leading solutions in application areas such as laptops, desktops, and clouds."

Arm's comprehensive computing solutions enable the ultimate digital experience

Although the global semiconductor industry has been changing in recent years, providing advanced computing power has always been the fundamental goal of industrial development. From this perspective, what challenges will chip design companies face? Zou Wei analyzed that the first is that the CPU load is becoming more and more complex, and the requirements will be different in different scenarios, and the challenges faced by each link are different; although Moore's Law has slowed down, the market's pursuit of performance and power consumption has not stopped, which is also a major challenge; in addition, the high price of masks and the longer production cycle make it very expensive to design a 5-nanometer chip and the production cycle very long. "We hope that chip design will not only cover current applications, but also future applications to obtain a longer life cycle, which requires us to respond to the challenges encountered by these customers through architecture, design, and IP." He emphasized.

If enabling the ultimate digital experience is the driving force behind Arm's vision for future computing, then Arm's latest comprehensive computing solution is the first step in the journey to the next decade in the Armv9 era. Arm's comprehensive computing solution adopts a system-wide holistic optimization approach across hardware IP, physical IP, software, tools and standards, providing Arm's partners with a wider range of choices to meet the application scenarios and cost ranges of all terminal market segments, allowing them to confidently use the latest technology to quickly bring high-performance products to market.

As the cornerstone of Arm's comprehensive computing solutions, the new Arm Cortex CPU includes Arm's most powerful Arm Cortex-X2, the first large-core CPU Arm Cortex-A710 based on the Armv9 architecture, and Arm's first high-efficiency small-core Arm Cortex-A510 and Armv9-A CPU cluster launched in the past four years. To support the ecosystem's demand for performance, Arm's global partners are working to ensure that all apps will support 64-bit by the end of this year. As Zou Wei said in his speech, in the era of diversified computing power, Arm believes that the computing power level of the CPU is still very important, and it has built the foundation for the digital future.

Behind the 20-30% performance improvement of each generation of CPU is Arm's unremitting pursuit of improving various indicators such as frequency, bandwidth, latency, cache, and process level, which has been driving the industry's progress in the innovation of computing infrastructure. Therefore, in addition to improving CPU performance, Arm's comprehensive computing solution also hopes to make system-level improvements to bring customers a better design experience.

In addition, visual experience is the key for consumers to interact with and enjoy devices. It is reported that the latest Arm Mali series GPU is being paired with the Armv9 CPU in a comprehensive computing solution to provide an excellent user experience. According to Zou Wei, in the future, the Mali series GPU will continue to exceed everyone's expectations, bringing ray tracing technology, variable rate shading VSR technology and other advanced rendering technologies.

Arm's interconnect technology is also critical to improving system performance. The latest CoreLink CI-700 consistent interconnect technology and CoreLink NI-700 on-chip network interconnect technology work seamlessly with Arm CPU, GPU and NPU IP to enhance system performance across SoC solutions. CoreLink CI-700 and CoreLink NI-700 provide hardware-level support for new Armv9-A features and support higher security, improved bandwidth and latency.

Introducing new security features to protect core data assets

As the number of connected devices surges around the world, the security of chips is becoming increasingly important. The fundamental need for security is to rethink how to approach security, and the Armv9 architecture redesigns the way secure applications work by introducing the Arm Confidential Compute Architecture (CCA). Confidential computing creates a hardware-based secure operating environment to perform calculations, protecting some code and data from being accessed or modified, even from privileged software.    

Zou Wei further explained that Arm CCA will introduce the concept of dynamically creating confidential realms (Realms). Confidential realms are for all applications and run in an environment independent of secure or non-secure environments. They are dynamic secure areas for storing data and executing code, separated from the privileged mode of the operating system or hypervisor to achieve the purpose of protecting data security. For example, in commercial applications, confidential realms can protect commercial confidential data and code in the system, whether they are being used, idle or in transit. The Arm confidential computing architecture is built on Arm's TrustZone technology and is a hardware version of the software container that allows applications to run easily on different systems. For example, applications running in non-secure areas can dynamically apply for confidential realms to protect their own algorithms or data.

"Arm also works with partners to develop the Platform Security Architecture (PSA), which is a set of standard threat models, measures and timelines provided to terminal designers. Combined with graded certification, it allows purchasers of this equipment to be more assured of the security of their assets. For example, in China, we have collaborated with China Academy of Information and Communications Technology Telecommunication Technology Laboratories to support large-scale deployment of secure IoT solutions based on the PSA framework," added Zou Wei.

Building a big computing platform for the new computing era

In addition to security, machine learning is also seen as an important part of Armv9, as more and more machine learning workloads will become commonplace in the next few years. In terms of machine learning, the Armv9 architecture supports the BFloat16 format to better support Int8 computing and BFloat16 machine learning; the introduction of Scalable Vector Extension 2 (SVE2) can better help developers develop high-level application scenarios, and has great gains when processing 5G, virtual reality and augmented reality, as well as image and speech recognition tasks.