Click the “blue text” above to select “Set as Favorite“
Key information, delivered on D1 time!
Arm has been working to expand into the data center market over the past few years, and these efforts are beginning to bear fruit. Several cloud computing providers, including the world’s largest AWS, are running Arm-based CPUs in their data centers. Last year, Fujitsu’s Fugaku supercomputer, built at its research center in Japan, utilized Arm’s CPUs and GPU accelerators, earning the title of the world’s fastest supercomputer.
If there are still doubts about Arm’s future development in the data center industry, those doubts will soon dissipate.
Arm Holdings is a CPU architecture design and development company headquartered in Cambridge, UK, primarily providing chip products for smartphones and IoT devices. Arm has been working to expand into the data center market over the past few years, and these efforts are beginning to bear fruit.
Several cloud computing providers, including the world’s largest AWS, are running Arm-based CPUs in their data centers. Last year, Fujitsu’s Fugaku supercomputer, built at its research center in Japan, utilized Arm’s CPUs and GPU accelerators, earning the title of the world’s fastest supercomputer.
In March of this year, Arm announced the launch of its next-generation Arm CPU, Armv9, which is expected to become a cornerstone product for the company in the coming years. Arm claims that this CPU will perform better in areas such as digital signal processing and machine learning compared to previous models, making Arm systems overall more robust and secure. However, compared to the processors from data center market leaders Intel and AMD, Arm’s chip products consume more power.
Arm hopes to launch more CPU products, depending on whether Nvidia can obtain the necessary approvals for its $40 billion acquisition of Arm from SoftBank. Even before the sale, Nvidia had already placed a significant portion of its data center product roadmap on the Arm architecture. Arm powers its BlueField SmartNIC, and its upcoming AI workload CPU, named “Grace,” is based on the Arm CPU architecture.
Applications of Arm CPU-based Servers in Enterprise Data Centers
Many people are skeptical about whether enterprises are deploying Arm servers in their data centers or cloud computing facilities. If they are, the numbers are negligible. Intel and AMD’s x86 architecture continues to dominate this industry.
Zachary Smith, General Manager of Bare Metal Business at data center operator Equinix, said, “I believe the application of Arm servers in data centers is still very limited. If we exclude those enterprises running on RISC architecture, using IBM mainframes and other computing resources, the vast majority of servers are still operating on x86 architecture. This has been the case for the past 15 to 20 years. And three or four years ago, it was difficult to find servers using AMD CPUs in data centers, although there has been a significant change since then.” He made this comment regarding the application of Arm chips in Equinix’s retail colocation data centers.
Other data center providers share similar views. However, they all believe that the adoption rate of Arm’s CPU products in data centers is on the rise.
Jeff Wittich, Chief Product Officer at Ampere, which designs Arm server CPUs for hyperscale and small cloud providers, said, “I would say we are just beginning to enter the market. If we go back three years, it would be fair to say that the adoption of Arm CPU servers in data centers was zero. Clearly, some Arm chips are used in SmartNICs, and perhaps in some edge devices and network facilities, but not in servers in data centers.”
Applications of Arm CPUs in Cloud Computing Data Centers
Currently, several major cloud computing providers are fully supporting the adoption of Arm CPU servers. Almost everyone is widely using Arm server CPUs behind the scenes to support their infrastructure, with some providing architectures for bare metal deployments or as infrastructure as a service. However, similar to on-premises data centers, most Arm server CPUs running in cloud computing data centers are used in devices like SmartNICs to offload network processing tasks from the CPU.
Vlad Galabov, head of cloud computing and data center research at Omdia, said, “Today, the primary use of Arm CPUs may be in Ethernet adapters or SmartNICs, and some cloud service providers have taken the lead in adopting them. But other markets are also adopting these products.
This allows network protocols, storage protocols, or any type of Security Device Manager (SDM) functionality to run on Arm processors in Ethernet adapters. This means that the CPU does not need to run the protocols, thus reducing the CPU’s workload and enabling faster data transmission. This may be the most widespread application today, as this market may be more mainstream than others.”
As more cloud server providers begin to offer Arm CPU-based servers to their customers, this situation has changed over the past few years.
Wittich said, “AWS has adopted Graviton2, and some Chinese cloud computing providers have adopted Huawei’s chips. From our perspective, we have been promoting our products to cloud computing providers since the second half of last year. Oracle Cloud has started adopting Arm servers. I believe we will see more announcements from large-scale cloud computing providers this year, launching some products and services, and all major infrastructure and service providers will have their own Arm-based instances.”
From GPU to Servers
Wittich stated that one of the reasons for excluding Arm servers from data centers is the lack of software that can run on that architecture.
He said, “In the past two years, some things have clearly changed. Arm has launched CPUs with significantly improved performance, and we are building a SoC around Arm’s CPUs that is actually designed for applications in data center servers and can connect to all the devices users expect to connect, such as GPUs and SmartNICs. Many software applications can run, but there are still some gaps.”
He noted that now all major Linux server distributions support Arm CPUs, and all major hypervisors can run on servers running Windows operating systems, with no gaps remaining.
Smith from Equinix stated that with the support of operating systems and hypervisors, the company’s bare metal cloud customers are becoming more willing to consider adopting Arm servers for workloads instead of sticking to x86 architecture servers.
He said, “Many of our customers have indicated that they do not care whether they adopt Arm or x86; it does not matter. What matters is which runs faster or costs less, or which server has more memory. This is where I see a significant shift happening. We even have some customers willing to adopt thousands of Arm servers.”
Applications of Arm in On-Premises Data Centers
In surveys, many believe that Arm servers will occupy a larger market share in on-premises data centers within two to five years, even if this does not fully align with their vision.
Smith believes that the adoption of Arm server CPUs is largely related to their flexibility, as device manufacturers can design Arm chips to meet specific needs rather than being forced to adopt Intel or AMD CPUs.
He said, “The trend of digital transformation will continue to accelerate, which means that people will see more companies with highly specific digital workloads emerging. These companies are committed to finding the right architecture for their software. Therefore, in five years, there will be more digital service providers choosing economically viable tools. Thus, I believe we will see a more diverse range of workloads and alternative architectures.”
He stated that this does not necessarily mean that the rise of Arm will come at the expense of Intel and AMD.
Smith said, “I also believe we will see traditional CPU manufacturers respond. We will soon see the emergence of new architectures, which is very fast in terms of server technology development. For example, Nvidia’s view on servers is completely different from Intel’s; Intel is CPU-centric, while Nvidia is data- or GPU-centric. I do not think we will see major changes in the next two years, but we will see more operational models. Five years from now, we will have highly specialized cloud platforms and those working on more integrated architectures.”
Galabov from Omdia stated that due to products like Nvidia’s Grace and ArmV9 promoting improvements at Arm, he expects a significant increase in the usage of Arm CPU servers by 2025.
He said, “Currently, we expect Arm-based servers to become mainstream. By 2025, their market share will approach 10%, but I believe that due to new market performances, development will accelerate. According to a vendor’s forecast, a significant portion of the global server market could be Arm servers by 2030.”
Galabov added, “If I had predicted five years ago that Arm would one day become a major server architecture, people would have thought I was crazy. But now, it is not surprising to predict that Arm servers could even surpass x86 servers by 2030. I am not making that prediction; I think it is unlikely, but it would not be surprising if it happened.”
Copyright Notice: This article is compiled by D1Net, and reprints must indicate the source at the beginning of the article: D1Net. If the source is not indicated, D1Net reserves the right to pursue legal responsibility.
(Source: D1Net)
If you work in a field of enterprise IT, networking, or communications and wish to share your views, you are welcome to submit to D1Net.
Submission Email: [email protected]
Click the blue text
to follow
D1net’s Xinzongzhi is an intelligence and resource sharing platform for CIOs (Chief Information Officers) and is also the largest CIO social platform in the country.
Xinzongzhi allows CIOs to serve CIOs, providing product reviews, consulting, training, headhunting, and demand matching services. It is also one of the earliest B2B sharing economy platforms in the country.
At the same time, D1net and more than half of the information department heads of central enterprises jointly established the China Enterprise Digitalization Alliance, mainly aimed at providing technical, policy, strategic, and tactical support for digital transformation to large enterprises across the country.
