Windows on Arm’s Last Stand? (Premium)

The consensus in the community is that Windows on Arm (WOA) is doomed, and that this platform will never go mainstream, let alone take over as the primary PC platform. And while I’m as clear-eyed as I can be on this topic, given my years of experience with WOA and the many WOA-based PCs I’ve tested, I do believe it still has a chance to succeed. And that’s true even if I remove emotion from the equation. Because truth be told, we need WOA.

In my Programming Windows series and the resulting book Windows Everywhere, I observed that the history of Windows can be seen as a series of reactions to competitors, market trends, and internal forces that were seeking change. Along the way, Microsoft worked to improve and adapt Windows to an incredible variety of new use cases, form factors, and competition. 40 years after its inception, Windows today would at least be familiar to those hardy few who used the first version, but I suspect that they would also be shocked both by its evolution and by the missteps that temporarily derailed progress along the way.

From my perspective as a Windows historian of sorts, I see many key milestones over the years. The moment when two low-level engineers rescued the otherwise abandoned product by figuring out how to adapt it for the 32-bit world, kicking off the escalating successes of Windows 3.0 and 3.1. Bill Gates’s decision to hire Dave Cutler and a team of ex-DEC engineers to create Windows NT. Jim Allchin’s stewardship in transitioning Windows to that NT codebase. The rise of the Internet and Microsoft’s bungled response with .NET and Longhorn. Dave Cutler’s adoption of the AMD x64 hardware platform as the way forward in the 64-bit world we are still within. The rise of mobile with the iPhone and the iPad, leading to the disastrous release of Windows 8. The pivot back to the desktop with Windows 10. And the Windows 11 push, which I can now tell you was a reaction to Apple Silicon and the Mac, and an era that’s not yet discussed in this series or book. (I will get to that in a future update.)

I’ve been holding that Windows 11 revelation in my back pocket, so to speak, but it bears on today’s discussion. What I can tell you is that it’s been confirmed by two highly placed sources, both former Microsoft executives: The software giant was so freaked out by Apple’s successes with the release of its initial M1 chipset in late 2020 and the ease at which it transitioned the user base and application ecosystem to this new platform, that it felt it needed to respond quickly. And so it threw together Windows 11 using left-over ideas from Windows 10X and threw it out the door before it was fully baked less than a year later. The theory was that it was better to ship quickly with mistakes and fix them over time than it was to wait until it was ready.

This is, of course, where WOA enters the picture.

The chipsets that power WOA-based PCs, like those made by Apple for the Mac, are based on the same underlying Arm architecture, which is a sharp departure from the traditional x86/x64-type systems that once powered the Mac and still power the PC. Microsoft has, of course, made two major pushes to bring Windows to Arm, first with the Windows RT derivative of Windows 8 in 2012 that was a reaction to the iPad and, as crazy as this now sounds, was expected to be what powered the volume seller in the Surface lineup. The second, of course, was Windows 10 on Arm, now Windows 11 on Arm, or just WOA, which was first introduced in late 2016, one year after the initial release of Windows 10.

I’ve spoken to then-Windows chief Terry Myerson on and off the record about this latest Arm push several times, and he was always clear that his anger at Intel was what drove this. Microsoft had been begging Intel for several years by that point to deliver more energy-efficient PC chipsets that could still perform acceptably well, but the microprocessor giant refused to move past its short-term gains in bulkier, inefficient chipsets with massive performance advantages. Echoing the “Megahertz wars” of the late 1990s, Intel introduced ever more powerful versions of its chipsets with no regard to efficiency over much of the past decade., offering only feeble Pentium- and Atom-branded chipsets in the low-end of the market and so-called U-series chipsets for portable computers.

But Microsoft—and Windows as a platform—needed more than that. Or, maybe less, if you’re looking at efficiency. With the rise of mobile devices, the software giant saw the next existential threat to its client platform coming in the form of iPads, Android tablets, and Chromebooks, each of which is more efficient on the same hardware than Windows. However, most of these devices do not run on the same hardware. Most of them run on even more efficient Arm-based hardware, platforms that combine excellent performance with a level of battery life that Windows PCs could never attain with their inefficient x64 underpinnings. For Windows to not just survive but thrive in this new era, Microsoft was forced to break with the “Wintel” traditions of the past and look elsewhere.

And it chose Qualcomm for better or worse.

To be fair, it made sense at the time, as Qualcomm was, and still is, the biggest maker of microprocessor SoCs (systems on a chip) in the world. Qualcomm had ridden to fame and fortune selling mobile communications chipsets—it was an early investor in CDMA—and while this is still a key focus today, most now know it more for its Arm-based microprocessor SoCs and its resulting dominance in mobile. In more recent years, it has taken the pragmatic step of dividing its SoC product lines according to the device types they serve, and so we see specific chipset families now for smartphones and tablets, IoT devices, wearable, and, yes, PCs.

But Apple’s rise during this time complicated matters. Though its mobile SoCs are used in more new devices each year than Apple’s, the Apple Silicon chipsets have routinely outperformed Qualcomm’s offerings (and it doesn’t help that Apple’s iPhone outsells Android overall now in the U.S.) and often by a wide margin. And where Apple successfully transitioned the Mac to its new M-series chipsets, Qualcomm has struggled with its non-smartphone chipsets. It just announced a switch to the RISC-V architecture for wearables, for example, an implicit admission that its current designs are a dead-end. And its PC-based chipsets, to date, have been lackluster, barely performing to the level of entry-level x64 hardware. Here, Apple Silicon maintains a huge lead. Indeed, that lead is so big that WOA runs better in emulation than it does on physical Qualcomm-based PCs.

This is all very embarrassing. More problematically, it suggests that Microsoft’s bet on Qualcomm was perhaps the wrong decision. And while Qualcomm enjoyed an 8-year exclusivity deal with Microsoft to supply WOA chipsets, that deal is running out at the end of next year. We’ve known for years that other chipmakers, including Samsung, have expressed interest in perhaps making WOA chipsets. And this week, we learned that AMD and NVIDIA will make WOA chipsets starting in 2025. Qualcomm’s monopoly of this thus-far tiny market is about to end.

Qualcomm has long understood its shaky position in the market, and it has made moves in the PC space to improve the performance of its chipsets. Key among them, of course, was its January 2021 acquisition of NUVIA, which had been designing a new type of Arm-based chipset for PCs with both high performance and the killer battery life for which Arm-based devices are known. Since this announcement, we have waited impatiently for Qualcomm to respond to the Apple Silicon threat, an effort that, if successful, would allow Windows 11 to compete head-to-head with new-generation Macs on a more even footing.

But it’s important to know that we’re not the only ones impatient with Qualcomm’s lack of progress. Microsoft has enthusiastically embraced the new companies waiting to supply WOA chipsets and among the many possibilities for the future of this platform is a diverse world of microprocessor SoCs from a variety of hardware makers and not the monopoly we see now in WOA and near-monopoly, with Intel, in the traditional x64 world. This change will benefit everyone up the food chain, from Microsoft to its customers. Well, everyone but Qualcomm.

With its exclusive access to WOA about to expire, Qualcomm today will unveil the NUVIA-based Snapdragon X Elite chipset, its last-ditch effort to remain relevant in the PC space. The firm faces hurdles of both perception and reality, thanks to it boldly claiming that it would deliver “Core i5” levels of performance in the past and failing horribly to do so. If the Snapdragon X doesn’t live up to the hype, it’s over. Maybe not for WOA, which could live to fight another day with AMD, NVIDIA, and probably others on board. But definitely for Qualcomm.

It was with this background in mind that I met with Qualcomm recently to find out what it had accomplished with the Snapdragon X Elite, the first model in a coming family of PC chipsets. On paper, it looks impressive. But then so did the previous generation Snapdragon chipsets for PCs. And so I was leery of the claims. I want to believe, of course, most of us do. But I’ve been bitten so many times in the past.

The Elite X’s 4-nm manufacturing process is unheard of in the PC space (its sort-of predecessor, the Snapdragon 8cx Gen 3 Compute Platform, was built on a 5-nm process, but most PC chipsets are much bigger and even less efficient). Its “Oryon” CPU runs at up to 3.8 GHz and offers a new 12-core design that departs from previous Snapdragon innards: These are all high-performance CPU cores thanks to its NUVIA architecture. Its Adreno graphics and Hexagon NPU AI engines are likewise impressive, on paper. It supports up to 64 GB of very fast LPDDR5x RAM (136 GB/s) and NVMe SSD over PCIe Gen 4 storage with UFS 4.0 support, also very fast. It’s compatible with USB4. And the communications capabilities are all leading-edge thanks in part to the Snapdragon X65 5G modem, with 5G, Wi-Fi 7, and Bluetooth 5.4 support.

This is all very reassuring.  But again, the past failures tug at me. And some of Qualcomm’s language around the Elite X troubles me.

The firm claims that the Elite X is “the most powerful, intelligent, and efficient processor in its class.” But what’s its class? Is it in a class of its own?

Where firms like HP routinely identify not just the firms that a particular PC competes with but their actual models, Qualcomm is being coy. It identifies the competition only as “the competition,” literally, rather than naming companies like Apple, HP, and Lenovo or actual products like the MacBook Pro. If it was so confident in its claims, wouldn’t it be specific?

Qualcomm mixes and matches between these competitors, too. It’s reasonable, for example, to expect the X Elite to perform somewhere in line with an Apple Silicon chipset and for the resulting PCs to deliver similar battery life. But when Qualcomm talks about matching the peak performance of a competitor while using just one-third the power, it is clearly comparing results to the x64 world only, not the Apple Silicon ecosystem. The Elite X “matches peak PC performance [of unnamed 10- and 12-core] laptop chips while using 1/3 the power,” it says. Is that a PC or Mac comparison?

(There is one instance in all this where Qualcomm goes after Apple Silicon in a reasonably obvious way: It claims that the Elite X offers “50 percent better peak multi-thread performance than [an unnamed] Arm-based competitor.” Since Qualcomm has no such competitors in the PC space yet, that has to be Apple. Right?)

Worse, many of its comparisons are with previous-generation Snapdragon chipsets, none of which were strong performers. So a “2X” faster CPU is somewhat meaningless. What’s 2X of garbage?

All this said, there are a few things in Qualcomm’s messaging that I don’t doubt. For example, with its rich history of NPU design, I’m sure its AI performance claims are at least generally true, and that anyone looking at this particular area will get the best results on Qualcomm: Intel and AMD are just spinning up their own NPU efforts, after all. But this is where Microsoft fails the relationship: It has yet to make a compelling case for on-device AI hardware on PCs and has never offered even a single example of an app or service that would benefit from this. (Beyond Windows Studio Effects, none of which is impressive.)

So what are we left with here?

Windows on Arm has failed to date, yes, and while that’s on both Microsoft and Qualcomm, it’s clear that Microsoft, over time, has done everything it can to fix or mitigate the platform’s software failings while Qualcomm has similarly failed to make commensurate gains with its hardware. Before the Elite X, Qualcomm made small gains in performance but they always came at the expense of battery life. The dream of these Arm-based systems, and the reality we see on Apple Silicon-based Macs, is the magic combination of both high performance and battery life. And so the question we’ve been asking for the past many years remains: Will Qualcomm ever deliver the home run we all need?

The proof will come from real-world usage. I wasn’t able to attend this week’s Snapdragon Summit, so it may be a while before I can get my hands on real hardware and see whether reality meets the marketing.

It hasn’t yet. But I still want to believe.