A Competitive Apple ARM Core Could Finally Break x86’s Long Computing Monopoly


Yesterday, reports surfaced claiming Apple will announce its first homegrown ARM CPU intended for desktop and laptop computing at WWDC in just a few weeks. Short-term, the announcement won’t change much, especially since Apple isn’t expected to launch actual ARM-based hardware at the event. Long-term, the impact on the x86 CPU market could be profound — or it could prove the truth of what AMD and Intel have been saying for years.

Note: In this story, “desktop market” should be translated as including desktops, laptops, 2-in-1s, and any other traditional devices that would normally run Windows, Linux, or macOS.

Eight to 10 years ago, pundits were predicting a giant collision of markets as Microsoft entered tablets with Windows and Intel launched a mobile chip business. Both of these efforts failed, and while Microsoft has a Windows on ARM product division today, the underlying hardware wins great marks for battery life and weaker scores everywhere else. There’s no evidence that these devices are lighting up sales charts. Intel, of course, ultimately left the smartphone and tablet market.

Instead of colliding, the desktop and mobile markets have continued along in their own tracks, with desktops remaining tied to the x86 market and the Intel/AMD duopoly, while ARM took over the mobile market. Apples-to-apples comparisons between x86 and ARM are difficult to engineer, due to profound differences in device types and market segments.

Launching its own mobile/desktop SoC will make Apple the only consumer hardware company on Earth that controls both its hardware and its software stack (unless you run Intel Linux, anyway). At the moment, Apple also builds the highest-performance mobile ARM CPU core you can buy. Its CPU cores have improved markedly over the past few product generations, while Intel has struggled to move on from 14nm. 10nm Ice Lake CPUs have been shipping for nearly a year, but evidently whatever roadmap Apple saw wasn’t good enough to persuade the company not to launch its own product efforts.

Process Nodes or Processor Architecture?

The question of whether CPU architecture X is better than x86 has been a popular one for decades, where X is understood to mean SPARC/MIPS/Itanium/PA RISC/POWER/ARM and every other architecture under the sun. Whatever conclusions academics might reach, x86 has certainly owned the market, vanquishing every competitive architecture from the PC industry. An Apple-based ARM SoC, however, is a threat to x86 in a way that Windows on ARM isn’t.

Power vs. Performance

From a 2014 University of Wisconsin comparison on CPU ISA power consumption.

The inevitable performance impact of emulation ensures that no ARM Windows PC is going to match the capabilities of an x86 system — not until applications are natively available for both chips, with equal levels of GPU support and low-level optimization. ARM’s Cortex CPUs, while capable performers, aren’t as powerful in single-threaded code as Apple chips — and given that we’re also talking about moving from mobile phone power envelopes to laptop and desktop power envelopes, Apple has very good reason to want the highest level of single-threaded performance it can get.

If AMD hadn’t launched Ryzen in 2017, we might frame the coming match-up as an evaluation of the question: “Does Intel build the best CPUs?” Given that AMD’s Ryzen is arguably the better CPU than Intel’s current Core family, however, I’d widen the scope of the query: “What matters more — processor architecture, or process node and engineering?”

The argument that x86 is at some kind of fundamental disadvantage compared with ARM relies on the idea that low-level features of the ARM ISA convey a substantial advantage over x86, or the fact that x86 CPUs translate native x86 code to internal micro-ops for execution. According to Intel and AMD, the power penalty for performing this kind of decode in hardware is tiny. Studies on the power efficiency of various ISA’s backed up this argument some years ago, claiming that above the microcontroller level, CPU design decisions like cache size, transistor usage, and other aspects of the physical design had a much heavier impact on power consumption than the ISA itself did.

If power consumption and performance are largely functions of design decisions, then process node and CPU architecture, not the CPU’s use of x86 or ARM, will control how fast and power-efficient it is. But if Apple can build ARM chips dramatically faster than x86 CPUs without relying on fundamental new approaches to computing, it could reignite the argument over replacing x86 with something else, at the exact same time it becomes clear that “something else” might provide a better long-term computing future. For decades, the engineers trying to make that argument had to make it in theory, against a reality in which x86 was the only game in town. With Apple throwing its full weight behind a custom implementation of the ARM architecture, that would no longer be the case.

A Windows on ARM system with a fraction of the performance of an Intel or AMD system but better battery life is an interesting alternative to a mainstream PC, but it isn’t a threat to the dominant x86 market. A first-class Apple laptop or desktop that demonstrates improved power efficiency and equal or superior performance would be a direct threat of a sort Intel and AMD haven’t faced in decades.

At first glance, the threat seems limited to Apple’s own PC market share, but I’m not so sure that’s true. Other chip vendors might decide to tackle the same problem, particularly if Apple wants to compete with x86 even in top-end servers. This, in turn, could spur interest in ARM-native Windows projects to a degree we haven’t seen yet. Microsoft already has the building blocks in place to enable a first-class ARM-native experience on Windows, so that’s taken care of. The more companies are interested in building a native ARM experience on Windows, the more effective Windows will be at competing against x86 when running on ARM chips.

For years, Intel CPUs were taken as the default “best” that the market was capable of. Since 2017, AMD has increasingly filled that role, with chips that offer far more CPU cores and much higher overall performance outside of narrow areas like gaming. Since actual Apple hardware isn’t expected to be available soon, I’m not going to try to speculate about how the products from all three companies will stack up, except to say the narrative would be a bit different if Apple beat one x86 company versus beating both.

If Apple proves capable of building a CPU that can match the best that AMD and Intel can build, it’ll demonstrate that ARM chips are capable of exceeding the limits of x86. That doesn’t mean consumers would drop them — I’d expect a slow shift at most, given Apple’s pricing and typical market strategies — but it would still be a shock to the PC industry. For the first time, a specific vendor would have a unique baked-in hardware advantage that no other OEM could match. It would light a fire under both Intel and AMD to remedy the situation.

I’m not saying I think this is going to happen, assuming the rumors are true in the first place. Even if Apple announces a highly competitive chip in a few weeks, it may not initially occupy a clear leadership position. Even if it occupies a leadership position, Apple might go its own way without attracting much interest from the rest of the market — or that leadership position might be eroded by the time actual hardware ships. Even if it occupies a leadership position, Intel and AMD might prove capable of matching or exceeding Apple’s improvements before the market began to shift in a significant way. All of these things are possible. But if Apple can challenge Intel and AMD on their own turf, in core desktop applications, it’ll be the end of an era in computing. x86 might still retake the performance lead long-term, but its decades-long unchallenged reign would finally be at an end.

Now Read:

Leave a Reply

Your email address will not be published. Required fields are marked *