How can Arm gain enterprise acceptance?

The Arm micro-architecture has enjoyed fabulous gains over the past few decades as its range expanded from embedded systems and mobile devices to cloud service providers deploying instances powered by their own brands of Arm-based chips. But enterprise use and acceptance haven’t materialized.

The Arm architecture was developed in 1985 by a team at Acorn Computers, a developer of microcomputers in the UK that released the first Arm chip, which was originally called the Acorn RISC Machine. Over time, Acorn’s computer business faded away, and rather than sell the processor, Arm — which was officially founded as a company in November 1990 as Advanced RISC Machines — sold processor designs that licensees could modify to create a unique design.

For the longest time, Arm was an also-ran that barely registered in any computing capacity except for mobile and embedded. Then came the fateful decision in 2007 by Intel to decline to design a custom chip for Apple for its planned smartphone. Apple instead went with Arm for the iPhone, and the rest is history.

Since then, Arm has become standard issue in every smartphone and tablet on the market. It has also found some success in low-end notebooks, particularly Chromebooks. But outside of SmartNICs, the enterprise remains elusive. It’s not from a lack of trying. There have been multiple efforts to sell Arm-based server chips, all of which have failed. These include:

• Calxeda EnergyCore: Launched in 2011, Calxeda was an early pioneer in Arm servers but failed because it was a 32-bit design, not 64-bit.
• Nvidia Project Denver: A rare flop for Nvidia, it gained no traction.
• Unnamed Samsung Arm server chips: Canceled before any public release.
• AMD Seattle (Opteron A1100) and K12: Both were ultimately abandoned as AMD shifted its strategy back to x86.
• Qualcomm Centriq: Qualcomm released the Centriq 2400, a 48-core Arm server processor, but ended production shortly after launch.
• Broadcom Vulcan: Broadcom developed a competitive chip but dropped it as part of a restructuring. The IP was later picked up by Cavium and later acquired by Marvell.
• HPE Project Moonshot: These servers made it to market, but HPE has since discontinued sales of them.

For its part, Arm has made significant effort to develop its Neoverse architecture. The Neoverse is a 64-bit design specifically for the server market. The N-series, now in its third generation, is targeted at enterprise servers doing general compute workloads, and the V-series, also in its third generation, is optimized and tuned for high-performance computing and advanced analytics.

The AMD effect

A key reason so many vendors tried to succeed with Arm is because at the time, in the mid-2010s, AMD was not a viable alternative to Intel. It was struggling and had serious performance lag. OEMs and customers generally don’t like to be locked into one vendor, and there was a desire for an alternative to Intel. That led many to turn their gaze to Arm.

But in 2017, AMD released the Zen architecture, which was equal if not superior to the Intel architecture. Zen made AMD competitive, and it fueled an explosive rebirth for a company that was near death a few years prior.

AMD now has about 30% market share, while Intel suffers from a loss of technology as well as corporate leadership. Now, customers have a choice of Intel or AMD, and they don’t have to worry about porting their applications to a new platform like they would have to do if they switched to Arm.

Analysts weigh in on Arm

Tim Crawford sees no demand for Arm in the data center. Crawford is president of AVOA, a CIO consultancy. In his role, he talks to IT professionals all the time, but he’s not hearing much interest in Arm.

“I don’t see Arm really making a dent, ever, into the general-purpose processor space,” Crawford said. “I think the opportunity for Arm is special applications and special silicon. If you look at the major cloud providers, their custom silicon is specifically built to do training or optimized to do inference. Arm is kind of in the same situation in the sense that it has to be optimized.”

“The problem [for Arm] is that there’s not necessarily a need to fulfill at this point in time,” said Rob Enderle, principal analyst with The Enderle Group. “Obviously, there’s always room for other solutions, but Arm is still going to face the challenge of software compatibility.”

And therein lies what may be Arm’s greatest challenge: software compatibility. Software doesn’t care (usually) if it’s on Intel or AMD, because both use the x86 architecture, with some differences in extensions. But Arm is a whole new platform, and that requires porting and testing. Enterprises generally don’t like disruption — which an architecture change would necessitate.

Arm’s cloud success

Where Arm has been successful is among hyperscale cloud service providers. Every significant player has made its own Arm-based custom CPU as a cheaper alternative to x86 CPUs. For example:

• Graviton series from AWS is used in a broad range of AWS EC2 instances and is based on Neoverse V2 cores, offering up to 96 cores, enhanced networking, and improved price-performance for enterprise workloads.
• Microsoft Azure Cobaltis based on Ampere Altra and Neoverse designs. It’s optimized for Microsoft cloud VMs, enterprise integration, Microsoft ecosystem workloads, and hybrid scenarios.
• Google Cloud Tau T2A instances are powered by Ampere Altra (Arm Neoverse) processors, optimized for cloud-native performance.
• Alibaba Cloud ECS g8y offers Arm instances based on Neoverse, targeted at web hosting, database workloads, and more.

That’s not to say that Arm has not had a victory or two. For the first time this past November, there were 18 Arm-based supercomputers on the top 500 list of supercomputers in the world. All 18 rely on the Grace CPU designed by Nvidia, which is built for high performance computing.

For standalone chipmakers, there are two options: Ampere Computing, which makes the Altra and Altra Max Processor specifically designed for cloud service providers; and Nvidia with the Grace processor, designed for use in tandem with its GPUs, which include the Hopper and Blackwell generations. Qualcomm is rumored to be trying yet again at a server chip, but nothing has been announced.

Arm’s strategy isn’t to try to take business away from the x86 market. Rather, it seeks to capture share in the cloud and eventually leverage that to get into the enterprise. It’s not trying to displace x86 where it lives; it’s trying to capture new computing areas that are mostly in the cloud where x86 has no dominance. From there, Arm hopes to work its way back into the enterprise, says Mohamed Awad, senior vice president and general manager of the infrastructure line of business at Arm.

“We think that our adoption, both in the cloud and on prem, continues to accelerate as their workloads advance and become more and more modernized,” Awad says. “Our focus has been about enabling [enterprises] for hybrid-type environments, and it’s really about as they move more and more of their software stack over to a cloud-based software stack- it becomes much more attractive to them on-prem.”

Arm’s efforts to capture share in the cloud through cloud service providers like Amazon and Microsoft has some runway. Awad says 80% of the Fortune 500 are on AWS infrastructure, and there are more than 70,000 customers using AWS Graviton. And at AWS:reInvent last December, AWS said 50% of all EC2 instances in the last few years were on Graviton.

But Enderle doesn’t think it will translate into Arm in the enterprise, mostly because Amazon has no incentive to sell hardware with its Graviton chip in it. “The whole point of hybrid is people who want to move workloads seamlessly between the environments, and you can’t do that with the separate processors,” he said. “Typically, you require the same platform on premise as you do in the cloud, so the migrations don’t result in breakage.”

Crawford said he has heard no interest from server vendors like HP, Dell, and Lenovo for a multitude of reasons. For starters, the customers aren’t asking for it. So, if the big server vendors aren’t building Arm-based servers, where do customers get them?

“I don’t think Arm has a place directly in the enterprise, or a product to sell specifically to the enterprise, because the enterprise is not going to create a custom platform themselves. They’re not going to build their own servers and then the operating system that sits on top of it,” Crawford said.

Awad said Arm is building the ecosystem around the processor similar to the way Nvidia has built a significant software ecosystem around its GPUs.

“If you look at what we’re doing is, our focus is on aspects of the software stack which allow developers to very quickly access whatever the underlying computer acceleration technology is that stuff like cloud AI,” he said.

And Arm may have been tossed a lifeline by Broadcom’s VMware, of all companies, which just announced plans to port its hypervisor to the Arm platform. However, this is targeted more at edge networking than data centers. Still, it’s a start.

In the end, Arm’s greatest challenge may be the reticence of enterprises to disrupt their servers and systems. Once you have a system operating smoothly, you don’t want to change things around without a very good reason, and while the price/performance argument that Arm has over x86 is considerable, enterprises may still be reluctant to completely turn over their data centers to a new architecture.