According to IDC, the data center market will grow from $597 million in 2020 to $1.3 trillion by 2025. At the same time, the global data volume will reach 175ZB in 2025. This is a test for the data center, and for the data center, the architecture affects the computing power and power consumption.
Both x86 and Arm are instruction sets (also known as ISAs, architectures). Instruction sets are a set of standard specifications for software and hardware. They do not have entities. Using this specification, corresponding processor chips and software systems can be designed. For example, the software is the nut, the hardware is the screw, and the instruction set is the size specification of the nut and screw.
The Arm architecture is gaining more and more status in the data center, and it will further widen the performance gap with the x86 architecture. Ampere Computing (Ampere Semiconductor, hereinafter referred to as Ampere) is a provider of a new generation of 64-bit Arm server architecture solutions. This company can provide extremely energy-efficient products.
Super performance sparks deployment boom
The cloud-native era has arrived. Gartner predicts that by 2025, cloud-native platforms will underlie more than 95% of new digital applications, up from less than 40% in 2021. In cloud infrastructure, the CPU is the foundation of the data center.
In cloud infrastructure, the CPU is the foundation of the data center. Under the huge market scale, if the traditional CPU architecture that has been used for more than 40 years is used, due to high energy consumption and low density, it is expected that by 2025, the energy demand will increase by 2 times, and the space needs to increase by 1.6 times.
Ampere is four years old and in the past 18 months, the 80-core Ampere Altra and 128-core Ampere Altra Max processors have been shipped to customers around the world. According to its 2022 annual strategy and product roadmap, Ampere will also release its self-developed Arm core CPU AmpereOne this year, which is based on 5nm and supports PCI gen5 and DDR5.
Ampere Altra and Ampere Altra Max cloud-native processors are highly attractive. Currently, more than 40 server platforms support the Ampere Altra series processors, and seven hyperscale data centers around the world are deploying Ampere products.
In the Chinese market, Tencent Cloud launched an instance based on the Ampere Altra processor SR1; Alibaba Cloud not only launched the c6r and g6r instances based on the Ampere Altra processor, but also launched a mobile cloud game service based on the Ampere processor; JD.com launched two An instance based on the Ampere Altra processor; UCloud provides customers with cloud products equipped with the Ampere processor.
In the global market, Microsoft launched the Azure virtual machine (VM) based on the Ampere Altra cloud-native processor; Oracle Cloud launched the A1 virtual machine and bare metal instance based on the Ampere Altra processor a year ago; Red Bull Racing, GenyMobile, University of Melbourne , University of Bristol and University of Jena, among others, are also benefiting from Ampere’s processors in their relevant cloud environment applications.
Why is Ampere so trusted? This is because Ampere’s cloud-native processors adhere to the design concept of being born for the cloud. In order to optimize power consumption and area, the characteristics that traditional architectures are not popular with the cloud are removed, and the single-core power consumption is more than 67% lower than that of traditional CPUs. Ampere’s cloud-native processors are also the most energy-efficient processors to run all general-purpose workloads, with the Ampere Altra family delivering 200% of its performance with 50% less power compared to x86 processors. In terms of AI inference, Ampere can provide cloud-native processors that surpass traditional x86 processors by 3 times and lead AI inference capabilities by nearly 4 times in terms of performance/watt.
Architecture is the key to the data center
Why does the Arm architecture provide such powerful performance for processors? Jeff Wittich, chief product officer at Ampere Computing, said that the biggest disadvantage of traditional x86 is that it is not designed for data centers, nor for cloud or cloud services.
x86 has been launched for a long time and has been used in the market for decades, but the original design and application scenarios of x86 are used in some mobile or non-cloud services. In the past 10 to 20 years, since there is no better choice in the market, x86 will also be used in cloud services.
Jeff Wittich emphasized that now Ampere Computing has launched the first cloud-native processor for cloud computing, focusing on cloud requirements and application scenarios, and has more advantages than x86.
As an example, Ampere Computing’s processors are very resilient to user interference and can withstand a very large amount of user interference. Both from a performance perspective and from a security perspective, you can scale with confidence and boldness without worrying about performance degradation as the number and pressure of workloads increase.
In addition, Ampere can currently support up to 128 cores, and this number will continue to increase in the future. Therefore, it has certain advantages in terms of the stability of the architecture and the performance of high-performance cores.
Differentiation and the future
When it comes to differentiation advantages, Jeff Wittich believes that Ampere has developed a general-purpose processor for cloud applications, which can be used in all workloads in the cloud, and can also provide customers with optimal performance, the largest number of cores The energy efficiency of the industry can ensure that users can achieve high scalability, high performance and low energy consumption during use. In addition, Ampere has a very broad, strong and global customer base, and cooperation with them can help Ampere develop more unique functions, which can be used in many different scenarios and provide unique value for different cloud architectures.
There are three pillars of future computing, CPU, GPU and DPU. The three pillars of computing have different roles. The CPU is equivalent to a housekeeper, responsible for thinking and processing various businesses; the GPU is equivalent to an artist, specializing in various image processing tasks; the DPU is equivalent to the front desk, responsible for packaging and unpacking various data. Ampere can provide the best general-purpose processing power, the CPU, while also cooperating with other types of computing. Ampere cloud-native processors will work with Nvidia GPUs for AI training and graphics rendering, or FPGAs, etc., on the same platform.
The Arm architecture has always been a shortcoming in the ecology, and Ampere will provide comprehensive support in the ecosystem and technical support team. Jeff Wittich emphasized that Ampere has more than 130 applications that are regression-tested on the website every day, including open source software and applications, and will update the code for test-related situations. At the same time, Ampere has also developed the Ampere Developer Program, hoping that more people, more users and customers will join the application ecosystem. Ampere is also continuously expanding the team of cloud service providers in China to achieve better service support.
Author: Fu Bin
This article is reproduced from: http://www.guokr.com/article/461607/
This site is for inclusion only, and the copyright belongs to the original author.