Arm Processors Gearing Towards Enterprise-Wide Adoption

The potential acquisition of Arm by NVIDIA for US$40 billion announced in September 2020 has once again brought the Arm ecosystem into focus. Historically, Arm has had huge success in power-constrained CPU for use in devices where power consumption is critical, or where the processing task is not hugely complex or varied, such as mobile devices, televisions, set-top boxes, and sensors used for IoT applications. Arm dominates these markets, facing little or no competition. On the other hand, Intel and AMD have enjoyed a similar level of dominance and success with their x86 CPU in the server, desktop, and notebook markets, but this dominance is about to be challenged.

Developments from major players including AWS’s Graviton2, Microsoft’s Windows 10 on Arm, Apple’s M1 chip, Qualcomm’s Snapdragon 8Cx platform, and NVIDIA’s Grace chip all hint at a future that builds on Arm’s success in their traditional market. AWS, in its testing of the Graviton2-powered instances, claims to have found up to 40% better price performance over comparable current-generation x86-based instances. Microsoft is known to use Windows Server on Arm-based chips for internal Azure processes, and the Apple M1 CPU has demonstrated that a customized Arm-based chip can deliver better overall performance than an equivalent x86 CPU. Elsewhere, Qualcomm claims its Snapdragon 8cx Gen 2 Arm-based chip has a distinct performance advantage over the equivalent Intel Core i5 processors; it boasts excellent power consumption credentials in addition. With Arm’s track record in heterogeneous computing and NVIDIA’s credentials in this area, could the market be setting up for a revolution and disruptive innovation?

The dominance of x86 architecture CPUs in the enterprise data center in recent times is undeniable, and while x86 manufacturers have not stopped innovating, a lack of external challengers has meant that the pace of innovation has been set by the two dominant manufacturers: Intel and AMD. The enterprise market is relatively stable in terms of growth and the powerful x86 architecture CPUs still hold their place in enterprise datacenters, but the landscape is changing. Workloads in general are becoming more heterogeneous, and this will lend them better to distribution. This includes the enterprise workload, and as such, there will be a growing number moving to public, private, and edge cloud environments. This growth market is ideal for adopting alternative technologies, disrupting, and innovating. Arm is at the front of that queue, and many are expecting it to make significant gains on x86 market share.

The decisions about what hardware to make available to the enterprise workloads in the cloud and at the edge are being made by the hyperscalers. When the comparisons are made at this massive scale, small differences translate into tangible cost and performance benefits to the customer. AWS has demonstrated, with its development of Arm-based processors, that it is not content to wait for the tech supply chain to innovate at its own pace. It wants to push the price to performance ratios for itself and its customers, and this is the hardware it will make available to those customers. Advances in hypervisor technologies have further abstracted the underlying hardware from the payloads being run in the cloud and at the edge; we expect further advances to have a subtle impact.

One of the barriers to the uptake of Arm technology in the enterprise is the perception that the Windows ecosystem is not yet ready. Microsoft is investing in Windows 10 on Arm. It is actively encouraging its developers to support their apps natively on ARM64 and late last year released x64 emulation to its Windows Insider Program. It is clear to the industry that Microsoft sees the operational value in running on Arm processors and has taken steps to ensure it is not left behind. If this push from Microsoft results in the ability for the enterprise to seamlessly transition their workloads between Intel and Arm processors, then we will see a swing to the use of the most cost-effective processor by the enterprise. ABI Research believes this could see Arm-based processors eating heavily into the market that x86 processors have thus far held almost exclusively.

ABI Research also expects new and potentially groundbreaking solutions as a result of collaborations between NVIDIA and Arm. Arm has a proven track record of working with heterogeneous compute; using CPUs with different properties to work together to produce a more efficient overall result is an old concept in mobile technology. Arm-based processors are having their barriers to widespread adoption challenged. This is evident from the work Microsoft is doing around making Windows 10 on Arm a success; the processors are also readily available to test with little effort in hyperscaler cloud-based offerings. This means that any new heterogeneous/accelerated technology will be more accessible to general compute workloads and could result in a strong uptake. For specialized compute workloads, the enterprise is more reliant on the software vendors to support their applications on ARM64. With the prospect of new heterogeneous solutions on the near horizon, it is time for the enterprise to apply pressure on those vendors to provide that support.

With the hyperscalers offering multiple platforms to fulfill the enterprise workload requirements, it is important that the consumer of those technologies takes the time to understand their own processing profile and workload. This could be as simple as isolating a regularly performed processing task and running it under controlled conditions across the different cloud instance types available, followed by a cost/performance analysis. Benchmarking to find the most cost-effective solution to processing requirements is key to an enterprise ensuring it is getting the best value for money from the hyperscalers. It should be performed as technology or workload changes; in today’s market, expect that to be more frequently.

The underlying hardware will increasingly become less relevant in the cloud and edge offerings. As the enterprise works more closely with the hyperscalers, they will have to relinquish a degree of control over the hardware, and they will be limited to what is on offer. It is important that the enterprise remains hyperscaler agnostic and strives for workload portability between suppliers so they do not get locked into using a narrow offering of compute architecture. Workload portability between CPU architectures will increase the number of options that are viable for an enterprise to select from the hyperscalers, further reducing the susceptibility to vendor lock in.

Both Intel and AMD are investing in x86 architecture, enterprise workloads are becoming more defined, and as such, the technology to efficiently process them is becoming more specialized. Complex instruction sets generate very capable and flexible CPUs; reduced instruction sets generate very efficient CPUs that can be more easily customized. Both architectures have a place; which one works best for the enterprise will depend on their processing profile. What is evident is that widespread adoption of Arm-based CPUs by the enterprise is becoming a concept the market should remain open to. What is also becoming evident is that the most common vehicle used to access the Arm datacenter will be public/private clouds and edge datacenters.