[spiralpolitik]

To understand Apple today you have to look to the past. Apple ended up being stuck with Motorola's inablity to deliver faster PowerPC chips. Whole product lines were delayed, or not possible. Effectively they gave control on when they could ship new product to a third party.

10 years later they are now in the same position with Intel. If Intel delays the next version of its product line by six months then Apple has to put things on hold. This is bad for a company like Apple as it could cause them to miss out on potentially lucrative periods (back to School, the Holiday season etc).

Ultimately I suspect in the very near term we will see Apple move off Intel, first for the laptops. LLVM IR would fit this strategy better than fat binaries as Apple would not have to wait until developers recompile. They can have the entire App Store available on day 1 of a product release.

[cwyers]

> 10 years later they are now in the same position with Intel. If Intel delays the next version of its product line by six months then Apple has to put things on hold. This is bad for a company like Apple as it could cause them to miss out on potentially lucrative periods (back to School, the Holiday season etc).

This article isn't about laptops, though. It's about ARM. Apple isn't dependent on anyone for ship dates in the ARM space -- they license the ISA, but they design their own chips based on the license. Yes, they're reliant on Samsung to fab the things, but Apple doesn't need their own ISA if they want to use their own fabs. I don't see how replacing ARM with a custom ISA helps Apple any.

(As for Intel -- it's not like Intel's other customers don't have the exact same sales periods that Apple does. So they're motivated. As others have said, if Intel slips deadlines like that, who's to say Apple has the ability to meet them where Intel couldn't?)

[WaxProlix]

Sort of ignorant and at least a little tangential here, but the notion of 'licensing' an instruction set seems a lot like licensing an API (cf the Java API copyright controversy) - are there parallels/precedents here or is it an unrelated issue?

[kragen]

The Federal Circuit's idiocy in the Oracle case aside, it's well established that copyrights don't cover methods of operation such as APIs or instruction sets. It's also well established that patents do, and that's what ARM licenses. Also they bundle mask works and copyrightable HDL code.

[WaxProlix]

Cool, thanks for responding - what is it about an ISA that makes it patentable where an interface might not be?

[kragen]

Interfaces are also patentable.

[hinkley]

It's a curious thing in business. People will do all sorts of things to keep from losing a big customer to a competitor. Some big companies keep multiple vendors and play them off each other. If I don't like you this week I place an order with the other guys.

Intel's weak spot has always (or at least often) been mips per watt. Apple has a whole slew of products in the ultraportable niche and given their focus could easily have more.

If a custom ARM-like chip gives them the option to stop using Intel chips on everything but the MacBook Pro line, that would be worth a lot to them, even if they continued to use Intel chips on those devices. Just the threat gets them concessions.

[seunosewa]

Intel CPUs have the highest performance per watt when running performance intensive code. The problem was that they could not scale down energy usage enough when performance requirements were very low, such as on mobile devices, and thats what they have been trying to fix for some years.

[AnIrishDuck]

> Apple isn't dependent on anyone for ship dates in the ARM space -- they license the ISA, but they design their own chips based on the license. [...] I don't see how replacing ARM with a custom ISA helps Apple any.

Well, what if the time comes when ARM cannot deliver processor designs with necessary improvements to power consumption or processing speed? If they hit that wall, a new processor architecture might be the only way forward.

Given Apple's resources and position, that kind of contingency planning makes sense to me. You can't exactly spin up a architecture design team overnight.

[Narishma]

Apple doesn't use ARM's processor designs. They just license the ISA and design their own CPUs.

[AnIrishDuck]

True, but the basic point still stands. ISA heavily constrains the final design on-die. ARM might not be able to deliver an ISA that fits Apple's needs.

Apple already has a tweaked version of ARM (ARMv7s) [1]. There may come a day where tweaks no longer cut it. At the end of another excellent post, from the same author [2]:

> If compiler–architecture co-design is on the table, much more radical opportunities are available.

Apple has both compiler and architecture teams. And if it sees ties to a specific architecture hurting its product development goals again? It seems like exactly the kind of company that would drop the ARM ISA completely.

Or at least like the kind of company that wants to keep that option on the table.

[1] http://www.linleygroup.com/newsletters/newsletter_detail.php... [2] http://adriansampson.net/blog/macroscalar.html

[tw04]

And if a company with as much experience and expertise as Intel misses their ship dates, what on earth makes Apple think they can do better?

I could see an argument for cutting cost by eliminating the margin that Intel dictates... but even that assumes that you can find a third party to manufacture the chips at a rate that is low enough to both cover the IP Intel has in their chips, and the manufacturing cost itself.

[Tloewald]

Well, first of all Apple could do it and fail. They're done that before. They're confident/arrogant/audacious enough to try things that they can't pull off.

Also, they could take a two-pronged approach -- cherry-pick talent from Intel, nVidia, etc. and offer them the opportunity to leapfrog baggage from the past. How much of the difficulty of moving the x86 platform forward is a result of the ludicrous amount of cruft? Grabbing a few of the smartest people and narrowing your focus to an easier problem gets you a long way.

Apple has managed to outpace the entire industry (including Intel) with its customized ARM cores (getting to 64-bit over a year ahead of everyone, and take a look at benchmarks between Apple's Ax CPUs and rivals usually running at far higher clocks with more RAM and sucking more power), and it got there by cherry-picking talent, omitting stuff it didn't need, focusing on design, and treating fabs as a commodity.

Looks like interesting times ahead.

[spiralpolitik]

Broadwell slipped by several months meaning that Apple couldn't do their traditional pre Holiday refresh in 2014. I believe the Skylake chipsets that would allow Apple to switch to using USB-C for the Pros aren't available until next quarter meaning it's possible that Apple could miss another fall refresh.

Given that the holiday quarter is Apples best season you can probably make bets that they are at least exploring their options.

[ceejayoz]

> And if a company with as much experience and expertise as Intel misses their ship dates, what on earth makes Apple think they can do better?

Intel has other clients who may be just as noisy about their own ship dates. They make custom processors for AWS, as an example. An Apple-owned subsidiary/division would have Apple as its sole priority.

[mtgx]

I don't think Intel sees anyone on the PC space as a major threat, so they aren't in a hurry to make everything perfect all the time.

[dlp211]

Except their bottom line. Anyone remember the cost to Intel for their faulty FPU in the P5?

[gherkin0]

> Except their bottom line. Anyone remember the cost to Intel for their faulty FPU in the P5?

I don't think that's as big of an issue with modern processors:

https://wiki.debian.org/Microcode:

> Processors from Intel and AMD may need updates to their microcode to operate correctly. These updates fix bugs/errata that can cause anything from incorrect processing, to code and data corruption, and system lockups.

[gsnedders]

Oh, there's still plenty of things beneath that level that can go wrong. Just look at Intel disabling HTM (through microcode, admittedly, but disabling a whole feature isn't fixing it to operate correctly) in Haswell/early Broadwell.

[bgibson]

I was thinking that too. But I wonder if Apple sees an opportunity to become the Intel of ARM chips - mobile and desktop, at least.

[wolfgke]

I don't believe that Apple has a tendency to sell their chips to other companies.

[benihana]

>And if a company with as much experience and expertise as Intel misses their ship dates, what on earth makes Apple think they can do better?

A decade of shipping millions of products that depend on very complicated supply chains on time and to great profit?

[dlp211]

Yes, because Intel doesn't do the same thing. /s

Granted, Intel's supply chains probably aren't nearly as difficult as Apple's, but the supply chain is probably the least difficult part of shipping the next generation of a chip.

[saosebastiao]

I wouldn't say that. From a materials and complexity standpoint, Apple has a much more difficult supply chain. But from a capital and planning standpoint, Intel has one of the most difficult supply chains out there. Every year they start building fabs that won't produce a single production chip for another 3 years, while planning on stuffing it full of capital intensive technology that isn't technologically feasible yet and might not even work. Practically speaking, Intel has to accurately forecast what Apple's, Oracle's, HP's, and Dell's sales are going to be in 5 years, a good 3 years before those companies even start their strategic and capital plans. A five year forecast for a yearly financial report at Apple might take a couple of economists a week to ballpark, whereas at Intel they probably have teams of economists, statisticians, etc. working around the year to improve their sales forecasts on horizons ranging from a week to a decade.

[kemiller]

I think the point is that Apple's in the same league or better, and it doesn't have to compete with Intel in a fair fight. Intel has to work with every x86 program ever made (more or less) and in dozens of different environments. Apple only has to optimize its own stuff, and can change its mind on a whim.

[chkuendig]

There's a big difference between Motorola and Intel: This time around all their competition is also stuck with Intel.

Switching to Intel guaranteed that they worst-case will always be on par with their competition. They'd lose that if they switch away to custom built silicon.

Now if ARM would start to make serious inroads in the Laptop/Desktop market that might change. But then again, thanks to P.A. Semi Apple has more than enough ARM knowledge already.

[ak217]

If Apple wants to move off Intel for desktop CPUs in the near future, good luck to them - the laws of physics apply to Apple, too. Intel has invested tens of billions of dollars in process technology, owns key IP and assets, and (after a long delay) is pushing the power envelope of their CPUs down aggressively, with their latest generation providing desktop-grade performance at 4.5W and more to come. Their delays may affect aggressive product roadmaps, but have more to do with the fact that Intel is not one but two process shrinks ahead of the competition, and are dealing with problems nobody else has seen before.

Intel is better than ever at the power-performance game, despite suffering badly from ARM. The performance gap is huge. Their current position is the result of a continuous sequence of design success and R&D reinvestment that started with Pentium M more than 10 years ago.

If anything, I expect Intel to finally start breaking into the phone market as they catch up on SoCs.

[arrrg]

Apple always want to keep all options open – unlike in the past – but that doesn’t necessarily mean Apple will move away from Intel. They probably have plans to do it and teams working on it, but as long as Intel can deliver what they need they won’t.

I think they are well prepared for every eventuality, but currently Intel’s and Apple’s interests are pretty well aligned.

Basically we are seeing a race towards a common goal from two directions here, though: Can Intel hit the low power goals (with sufficient performance) before Apple hits their performance goals (at low power) through some alternate route? Honestly, I don’t see Apple being able to touch Intel’s performance and with regards to power use Intel is getting better all the time. The space for ARM-like laptop chips is getting smaller by the day.

And Intel actually wants to hit that low power! It’s not like they are disinterested in that (like IBM being disinterested in making the CPUs Apple needs back in the day because they were making their money elsewhere and Apple was just a small, unimportant customer of them, not worth all the effort), they are working on that all the time.

That’s what I mean when I say their interests are aligned. I mean, I obviously think that Apple is always on the lookout for alternatives (and they have been through many such transitions by now, so I’m very confident that they would be able to pull it off), but I honestly think they would prefer their and Intel’s interests to remain aligned and Intel just making some kick-ass low-power CPUs, exactly what they need for their future retina resolution, light-weight, super-thin, all-day battery life, fanless MacBooks.

[rythie]

I really doubt Apple is going to do anything radical like that with it's Intel based products. To do so would be massively expensive, dwarfing the revenue from it's Intel-based products, let alone any benefit. Apple's Intel based products are a very small part it's revenue and have almost no growth.

If they do this, it will be for it's ARM products, iPhone, iPad, iPod touch & AppleTV.

[cmrdporcupine]

Yep, and a victim not just the PowerPC nightmare but two other minority architectures _before_ that, both the 65xx and the 680x0. Arguably they didn't care as much about the 6502/65816 because they were already moving to the Mac/68k across their whole platform (people often forget that the Apple II line was Apple's major product and revenue source for their first 10 years), but in the early 90s Apple, Atari, and Commodore were all left in the lurch by the decline of the 68000 architecture. The PowerPC was the nominated successor, but it dragged Apple through 10 years of transition, with buggy backwards compatibility, porting an OS that was never designed to be portable, with major sections of it running as emulated 68k for years after they stopped shipping 68k machines...

[brudgers]

M68000 wasn't a minority architecture at the time Apple adopted it. It was a workstation workhorse. That's part of what justified the Lisa's $10,000 price in 1983. But a lot of it was margin and hence a year later 68000 Macs could be sold for less than a quarter that price but still offer incredible performance on a personal computer. By adopting a *nix based OS, Apple reduced most hardware compatibility issues to recompilation, performance of course being another matter.

As for the dark days of Apple in the PC market and the deaths of Commodore and Atari, my recollection is that it had more to do with the stock market crash of 1987 tightening access to capital and the S&L crisis that followed it creating a recession where downsizing was what corporations not empty-nesters did.

[cmrdporcupine]

At the time they adopted it but soon after. The moment the 386 became affordable the 68k platform was in decline.

[bryanlarsen]

If the 65816 (or something very similar to it) was available in 1978, the computing world would probably be very different now. 1978 was the year that both the 8086 and 6809 were released, which were similar 8/16 bit chips. 1979 saw the release of the 68000, which was a much more powerful chip than 1983's 65c816.

[cmrdporcupine]

68000 was more powerful for addressing large amounts of RAM that nobody could afford. In terms of actual workhorse performance there's a good argument to be made that a similarly clocked 65816 could outperform a 68000, unless your benchmark is heavy on 32-bit integer calculations. Instructions per cycle and interrupt responsiveness are higher for the 65xx (and 6809)

I'd prefer a 6809 with a wider address bus over a 68000. I wish Motorola had improved on the 6809 rather than having two competing architectures. The 6809 is a joy to code for.

[wiremine]

I understand the business logic, but as a non-hardware guy, I've often wondered at how this would look in reality:

- Apple has proven they can switch out hardware stacks, so that part seems straight forward enough. Especially given how mature XCode is.

- But: can they really get to a point where they complete with Intel directly? Would they try and use ARM or roll their own ADM64?

[JustSomeNobody]

At this point do they HAVE to compete with Intel? The other thing Apple is REALLY good at is telling people what they want. Apple would likely be able to convince enough people that an ARM base laptop is perfect for them.

[jsprogrammer]

Most Apple customers have no idea what ARM or ISA mean. Apple's not big on publishing detailed specs, so this doesn't even seem like a marketing issue. Apple will just put it in and people will buy it based on Apple's reputation alone.

[JustSomeNobody]

Excellent point.

[jsprogrammer]

I wonder how many Apple customers were even aware of the PPC->X86 switch several years ago, let alone how many customers still remember it?

[wolfgke]

> At this point do they HAVE to compete with Intel? The other thing Apple is REALLY good at is telling people what they want. Apple would likely be able to convince enough people that an ARM base laptop is perfect for them.

In my opinion it was mainly Steve Jobs who was brilliant telling people what they want. So I personally doubt whether this strategy still works with Tim Cook as CEO.

[rarepostinlurkr]

And in all honestly, they are probably correct in telling people that.

People are already using incredibly low power/perf laptops (new MacBook).

I'm not convinced anyone doing content consumption would notice if the chip swapped out for an ARM instead. After all, ARM's already power their content consumption devices (all of iOS)

[nsxwolf]

That would be a bummer for me and a lot of other developers I'm sure. I know we're not the target market, but I like using Apple hardware and OS X, but for work I am always running multiple VMs with Windows and Linux.

I'd be sad if my upgrade path became some big, noisy, inelegant HP Xeon box.

[leejoramo]

I would add that the NeXT Computer was faced with similar problems, and had to switch between different processors several times.

Apple/NeXt has much experience with technical challenge of porting entire platforms to different processors and with the business challenges of being tied to critical technologies you do not control.

[kjs3]

NeXT used the 68k from inception to purchase by Apple. While they tinkered with 88k and PPC, they never shipped those.

[fanf2]

They shipped NeXTstep for x86, SPARC, and PA RISC too.

[kjs3]

Very true, but the post I responded to claimed NeXT "had to switch between different processors several times", which isn't correct.