Intel and 10 nm (Premium)

Intel has always had a tortured relationship with mobile computing. But its inability to make its chipsets more efficient is embarrassing and troubling.

As part of its otherwise stellar earnings announcement this past week, Intel admitted that it was delaying its move to a more efficient and mobile-friendly 10 nm manufacturing process to 2019. This isn’t the first delay, either. Originally announced as a 2015 product, Intel’s 10 nm CPUs, now codenamed Cannon Lake, have been delayed multiple times.

Intel’s inability to reach 10 nm manufacturing in volume—the CPU giant inexplicably claims it is now shipping 10 nm chips in “low volume” without explaining to whom or for what purpose—has been an ongoing embarrassment for the firm. And it has done the equivalent of hand-waving to distract us from this fact by squeezing ever more performance—and, more important, more efficiency—out its current 14 nm process.

Last year’s release of the quad-core U-series chips, which hit at the mainstream part of the market, is the latest example. And the effort has paid off: Intel is providing a meaningful performance boost, essentially for free, and it is doing so without impacting battery life when compared to previous-generation dual-core chips. It’s a win-win, right?

Not exactly. The quad-core U-series CPUs are indeed a win for the shrinking audience of PC buyers who are upgrading here in 2018, for sure. But the current mass market for personal computing devices has moved past the PC to smartphones and tablets. And that much bigger market—smartphones alone now outsell PCs by a factor of at least 6 each year—is driven by more efficient CPUs and chipsets based on ARM.

Those ARM chipsets have a few meaningful advantages over Intel’s offerings. One is architectural: Rather than provide multiple physical and virtual cores, each with the exact same characteristics, ARM chipsets are optimized for thin, light, and tiny mobile devices. They have some combination of performance efficiency cores which let devices built on this platform run at full speed while used interactively. But they operate more efficiently while sleeping or unused, too, and provide dramatic battery life and standby advantages over Intel designs.

More to the point, the ARM world has already moved to the smaller and more efficient 10 nm manufacturing process that still eludes Intel. Qualcomm, the biggest maker of ARM chipsets, introduced its first 10 nm chipset, the Snapdragon 835, in 2016. It went on to be a best-seller, and was used in virtually every handset flagship (besides Apple’s) over the subsequent year. And some ARM makers are now working on 7- and 8 nm processes.

The ongoing improvements to Intel’s 14 nm chips are certainly laudable. But one gets the feeling that this effort has been forced on the company, and that it would have begun offering even more efficient 10 nm chips for portable PCs by now if it could do so.

This failing also raises the issue of what Intel is really competing with here. In the PC space, AMD has seen a small resurgence in recent years thanks to its own chipset improvements. But AMD still remains a minority player, is still “circling the drain” as one Microsoft insider told me recently. The real competition here isn’t AMD, it’s ARM. And yet the improvements that Intel has been able to make only address the AMD threat, such as it is. Intel is in a holding pattern and it is not addressing the ARM threat.

Intel’s saving grace, such as it is, is that ARM simply makes no sense today in the PC space. My own testing of Windows 10 on ARM shows that the performance issues simply aren’t overcome by the battery life and standby time advantages that this platform provides. ARM simply makes no sense for PCs today.

That buys Intel some time. And I think it explains this week’s delay announcement.

Hot on the heels of a well-received family of quad-core chips and a stellar financial quarter, Intel is awash in good vibes. And the first generation of Windows 10 on ARM PCs has arrived with a thud. Actually, less than a thud. It’s been pretty much silent.

So we’ll get yet another generation of 14 nm chips from Intel, codenamed Whiskey Lake. And it will have to ship 10 nm chips at some point, right?

Maybe not. Based on Intel’s statements this week about the difficulty of moving to a manufacturing process that can reliably handle the density of 10 nm chips, I’m starting to wonder if Intel will need to make a bigger change. This could involve a major overhaul of its manufacturing process that could trigger even more delays. Or maybe it just gives up on 10 nm versions of its current architecture.

Consider the impact of the Spectre and Meltdown on chipset design, and that the basic architecture of Intel’s chipsets simply do not meet the needs of modern personal computing devices. It’s possible that the firm will need to start over, or create a new processor family side-effort, to address these shifts.

Like Microsoft, Intel has weathered industry shifts before, in its case with such things as the change to the Core micro-architecture and its adoption of AMD’s 64-bit instruction set, which Intel brands as x86-64, or x64. So there is reason to believe it can do so again.

But like Microsoft, Intel may find itself relegated to dominating a minority subset of a personal computing market that it helped create. If all it can do is improve its current chips in iterative ways, it may find itself watching the industry move forward without it.

I think that’s already happening. That Intel’s role continues diminishing over time. And that we’re watching a potential extinction moment play out in real time.