I realized something interesting this December. For the first time in as long as I can remember, I don’t own a single x86 computer. In fact, the only thing with an x86 processor in my house — at least that I know of — is in a dusty, old Xbox One hidden away in my entertainment center.

Late this fall, my fiancé and I traded in our aging MacBooks for a couple of Apple’s new Arm-based M1 Macs. A base model MacBook Air for her and slightly upgraded Pro for me.

Ever since Apple announced it was putting an end to its nearly 15-year marriage to Intel this summer, I’d been patiently waiting to see how well Apple’s first crack at an Arm Mac would go. So far, it looks promising, but not without challenges.

Now this isn’t a review of either machine, but what I will say is my experience thus far hasn’t been nearly as critical as Moor Insights and Strategy’s or as glowing as The Verge’s. Battery life on these machines is good, performance is better, and most importantly, all of the software I use every day worked without issue.

That last bit is an important caveat. Most software isn’t running natively on these machines yet. Instead, apps written for Intel Macs run through a translation layer Apple calls Rosetta 2. At least for me, it just works. Photoshop and Lightroom are both fast, and Zoom and Webex haven’t given me any trouble. But your mileage may vary. Patrick Moorhead, president and principal analyst at Moor Insights and Strategy, reported numerous software issues in his review.

An Allegory of What’s to Come?

This mix of software and hardware is as critical for making Arm work in personal computers as it is in the data center and the cloud. However, like with the Mac, the software burden in the data center appears to be less an impassable chasm as a bump in the road.

And as far as the hardware goes, Ampere’s SVP of Product Jeff Wittich argues that the biggest hurdle for Arm in the data center isn’t a technological one, it’s an ecosystem one. It’s no longer about can you build a fast and efficient chip, it’s about convincing people to use it and write code for it, something that Arm itself has worked dutifully to make easier.

The Apple M1 is a “watershed moment” for Arm in the client space and an allegory for an inevitable shift to Arm in the data center, Wittich said, in an interview with SDxCentral. Apple’s M1 will no doubt be a catalyst for capturing mindshare and helping to move developers away from x86 to Arm, he said. “Now they’ve got a machine that’s in their laps that they can go and write code on.”

The New Kids on the Block

Ampere has a vested interest in the success of Arm in the data center. The company is among the first to produce an Arm-based server chip that is generally available outside a handful of public cloud offerings from the likes of Amazon and Microsoft.

Last winter, the company announced an 80-core data center chip called Altra, and this summer it followed up with an even denser version with up to 128 cores called the Altra Max.

However, Ampere isn’t the only horse in this race. Both Marvell and Amazon have invested heavily in building out Arm-based chips for the cloud.

This spring, Marvell unveiled its third-generation ThunderX chip. The 7-nanometer processor will offer between 60 and 96 cores, depending on whether a single or dual die socket is being used. Each of those cores offer four threads — up to 384 on the dual-die socket — thanks to 4-way surface mount technology.

According to Rabin Sugumar, lead architect for ThunderX3 at Marvell, the new chip offers a massive improvement in performance — up to 300% over the previous generation ThunderX2 — in highly threaded workloads like heavily populated MySQL databases.

The pursuit of higher performance per watt appears to be driving much of Arm’s growth, and if you want evidence just look at Fujitsu’s Fugaku supercomputer. The Arm-based system sits atop both the Top 500 and formerly led the Green 500, which ranks supercomputers by performance per watt. While still the performance champ, it has since fallen to the No. 10 spot for efficiency.

Seeing the success of Arm in high-performance compute, several cloud providers including Amazon, Microsoft, and Oracle have moved to adopt or develop their own Arm-based processors.

In early December, Amazon Web Services CEO Andy Jassy signaled the company’s commitment to custom silicon, announcing that the company would be shifting more of its public cloud offering to chips like the Graviton2, which launched this spring.

Graviton2 is the company’s second internally developed general-compute processor. It’s based on Arm’s 64-bit Neoverse microarchitecture, and is built using a 7-nanometer manufacturing process. “We realized a few years ago if we wanted to continue to push the envelope of price to performance, we knew we were going to have to develop some of our own chips,” Jassy said.

And while Jassy reassured Intel and AMD that their chips weren’t going anywhere soon, the message was clear: the future of AWS would be paved not on legacy architectures but purpose-built silicon.

Amazon doesn’t appear to be the only cloud provider that believes this either. In mid-December, Bloomberg, citing anonymous sources within Microsoft, reported the software and cloud giant was developing an Arm-based data center chip of their own to compete against Amazon.

Competitive Chip Landscape

Apple’s divorce of Intel probably won’t have a demonstrable affect on Intel’s profits. It certainly doesn’t look good, but Macs only account for 14.5% of all PC sales, according to Statista. By comparison, Intel owns more than 60% of the PC market, while AMD claims a healthy 37.3%.

And in the data center space, estimates put Intel’s marketshare somewhere in the low 90% of all servers sold. If you just look at marketshare, Intel is the obvious winner, but also has the most to lose.

It’s only when you start taking the company’s product roadmap and that of its competition into consideration a trend starts to emerge.

While Intel may be the top dog for now, its technological lead has all but eroded in recent years. The company’s flagship Xeon Scalable processors, even in their third-iteration, are still based on a now 6-year-old, 14-nanometer manufacturing process. The company isn’t expected to launch its first 10-nanometer Xeons until sometime in 2021, and its upcoming 7-nanometer process has been delayed until late 2022 or early 2023.

By comparison, AMD’s now year-old EPYC 2 processors, code named Rome, are based on Taiwan Semiconductor Manufacturing Co.’s 7-nanometer manufacturing process. This has allowed the company to produce chips with as many as 64-cores, more than twice as many as Intel’s highest specced Xeons.

However, the best evidence of this growing technological divide can be found in the two company’s financials. The past two quarters have been difficult for Intel, as the company continues to fight slipping deadlines and now declining data-center revenues. During Intel's third-quarter 2020 earnings, the company reported a 4% year over year decline in revenues and a nearly 30% decline in net incomes. By comparison, AMD has seen its revenues climb 56% year over year, and its net incomes more than double, during the same quarter.

Existential Threat or ‘Experimentation in the Cloud?’

While Arm still represents a microscopic sliver of the data center and cloud market today, that could change quickly, according to a recent IDC report.

“Worldwide revenues for servers running AMD CPUs were up 112.4% year over year, while ARM-based servers grew revenues 430.5% year over year, albeit on a very small base of revenue,” said Paul Maguranis, senior research analyst, infrastructure platforms and technologies at IDC.

Despite growing momentum, Dell’Oro research director Baron Fung doesn’t expect Intel will lose much ground, at least not in the new year.

“We project Intel to retain a strong leadership position going into 2021. Intel still has a commanding share among the top 10 cloud service providers, and this is a market that will undergo an expansion in 2021 with the ramp of the new Intel Ice Lake processor platform,” he wrote in a recent blog post.

But while Fung doesn’t believe Intel is at risk of losing its throne yet, he does expect AMD will continue to gain marketshare and for Arm to begin carving out a home in “niche markets and applications.”

Intel’s confidence in its ability to gain marketshare — not only in the CPU space but in the GPU and FGPA markets — was heard loud and clear when the company issued an unsolicited and arguably dismissive statement in response to Microsoft’s rumored data center chips.

“The incredible demand for computing fueled by new workloads like [artificial intelligence] is driving more silicon experimentation in the cloud,” an Intel spokesperson said in a statement provided to SDxCentral. “Building on decades of x86 ecosystem innovation, we are committed to providing customers the world’s best CPUs and new products from GPUs to AI chips. In this expanding market, we expect to gain share in many areas like AI training, 5G networks, graphics, and autonomous driving.”

Yes, Arm’s marketshare in the data center is absolutely tiny compared to Intel or AMD, but to call Ampere, Marvell, or Amazon’s efforts “experiments” is somewhat telling. Why would cloud providers like Microsoft, Amazon, or Oracle be “experimenting” with these chips if they didn’t offer something they couldn’t get with Intel?

More to Come in 2021

Things could get even more interesting if Nvidia can overcome the regulatory hurdles required to complete its $40 billion acquisition of Arm. The company has expressed an interest in expanding the development of the Neoverse microarchitecture used in Graviton2, ThunderX2, and Altra. What’s more, these same chipmakers could soon license Nvidia’s own GPU intellectual property and integrate it into future designs.

Arm hasn’t been sitting still either. Early this fall, the chip designer unveiled its updated Neoverse architecture which it promises will boost performance 40% to 50% over the previous generation N1 design, while consuming the same amount of power. The company also detailed a new Neoverse variant designed to go toe to toe with Intel in single-threaded workloads.

And Arm expects Neoverse chips to find homes in more than just CPUs. The company is targeting a swath of use cases like SmartNICs, cloud, enterprise networking, and at the edge where per-core performance is secondary to core count or energy efficiency.

Suffice to say, even if Arm doesn’t overthrow Intel or AMD in the next 12 months, it certainly isn’t going away anytime soon.