Apple to ditch Intel – report

Isn't Windows running on ARM these days too?

I think I heard something about some Windows ARM tablet being sold by some big name company whose name describes everything most women don't want on a penis.

You seriously overestimate Mac users' interest in Windows

I assume that this is a troll, but... seeing as Dassault Systèmes (Solidworks) estimates that only a fraction of one percent of OSX machines have installed Windows (any version), I doubt if Apple is losing any sleep over it. Recent versions of Boot Camp have already dropped support for anything but Windows 7 because of a near total lack of interest.

I find it highly doubtful myself but there are some intriguing possibilities. Microsoft is in business to sell Windows licenses more than anything else. If they cut a deal with Apple to allow a special version of Windows RT for Boot Camp that allowed desktop apps under x86 emulation, this speed things up considerably by having the OS and APIs be in native code. Sort of like WINE but with an emulator.

Remember the emulator for running x86 Windows apps on the DEC Alpha under NT/2000? It was called FX!32 and had a remarkable feature unlike other emulators: it saved the translated code produced by the emulator. This meant the app's performance increased with use as more and more of it became native and optimized. (Some hints suggest Microsoft used something similar in the emulation of the original Xbox on the Xbox 360, which is why a download is needed when an old Xbox game is run for the first time on a 360.) In this day and age most popular apps could have existing translated code downloaded automatically, greatly speeding up the process.

If Apple and Microsoft collaborated on such a scheme Microsoft would of course retain the rights to use the technology on other ARM platforms, after a period of exclusivity for Apple, offering a path for Windows RT to be leveraged further against Intel the next time Microsoft needs something Intel doesn't want to give them.

I guess it's always possible that when those "people familiar with the company's research" say that Apple is "exploring ways" to use its own silicon in future Macs what's actually being proposed is the consolidation of every non-CPU function into a single chip, keeping the Intel processor and moving to a [pretty much] two-chip design? They're already doing soldered RAM and SSD, the iPad designs are triple-sandwiched CPU+GPU+RAM chips and peering at the current Macbook Pro motherboard on Google images appears to show tens of chips across the board.

They're already using the on-board GPUs but I guess that building RAM+SSD into a single unit would be a good saving? It doesn't feel like something that's likely to come onto the market from anyone else.

Re: research

This comment is far too sensible for this thread.

Other topic: power use. I bet Apple would love to get even closer to making the computer disappear than they have so far.

Re: research

That was the feeling I was getting from the article: that a move away from Intel would be something Apple would do like five years down the road IF conditions were favorable. This sounds more like back-burner stuff: stuff to consider down the road, not stuff that's on a priority track. Jumping to PowerPC and then to Intel in the past made sense since both processor lines were stalling in the face of continuing x86/x64 development. Now processor development IN GENERAL is starting to stall because of physical limitations (processors can only be so small and go so fast), and R&D is now turning towards finding other ways to improve performance. ARM has shown tremendous progress in the portable market, but as the article notes it's had trouble penetrating the performance market because arm's energy-efficiency focus makes it less suited for high-performance applications.

Re: research

I can imagine them integrating ARM into the current Intel motherboards to allow a splashtop style iOS for Mac laptops, trading performance for massivly extended battery runtimes.

Are they going to roll their own x86 though? it would be quite expensive.

Anyway, one thing is for sure it is a lot easier to run CISC code on a RISC type design than the other way around. You can assemble RISC instructions into equivalent CISC instructions. But of course, byte ordering is also important. If your emulator spends half of its time flipping byte orders around then it won't run too fast.

The PPC chip had some sort of switchable byte order to get around the above problem.

Different corcumstances

In both previous architecture shifts the new CPU family was a major boost over what the previous machines had. For instance, at the time the PowerPC 601 was first shipped (although it was not intended to be a production CPU and had been built as a proof of concept) it was the fastest microprocessor BYTE magazine had ever tested. A major jump for that era.

The PowerPC models Apple used also had the advantage of a dedicated bit of transistor real estate to do some of the conversions that would otherwise have added a LOT of overhead to emulating the Moto 68K family.

When it was decided to go to Intel the problem was that Apple alone wasn't a big enough customer for IBM to commit the resources in producing competitive desktop CPUs at a pace to match Intel. The profit margins were tiny compared to IBM mainframes but the level of capital required was greater. Apple had no interest in encouraging an open market for non-Apple PowerPC desktops, so something had to change. By the time the first Intel Macs shipped the last round of CPUs IBM produced for Apple were getting a bit dated by the standards of the PC industry. Once again, the new architecture had plenty of spare horsepower to handle the emulation problem.

Unless ARM's upcoming 64-bit product line offers unprecedented level of performance gain, at least an order of magnitude over current ARM designs, there is just no way to have a third relatively easy transition. Something has to give. Either Apple dumps its professional users and allows their desktops to age into uselessness, or they once again offer some hardware means to ease the transition.

I'd expect the former to occur as those users have become a very small portion of Apple's revenue picture. The cachet of holding that market is no longer as useful as it once was now that they've become so strong in the consumer end.

Another possibility is that Apple continues on x86 but dumps Intel. AMD is currently circling the drain but has some very valuable IP. Apple could buy the lot using the change under the cushions of the couch in Tim Cook's office. One question would be whether to retain the ATI portion of the company or spin it off into an independent company again.

It has entertained this idea

Repeatedly. Windows 2008 for Itanium, Windows RT for ARM, and so on.

Those products were minority interests, or (rumour has it) underpowered crap, but that's not the point: Microsoft aren't loyal to Intel per se, they go where the market's going.

And Apple fans did take it, in 2005-6, and liked it, because it got them out of a processor dead end. I don't hope that Apple do this, because I think it would be far preferable and simpler (for users, that is) to try to develop lower-voltage x86 processors, but that's not the point. Apple was offering something extra to make up for the inconvenience: much faster processors. What if Apple now offered users iPad-style battery life on their laptops? I'm not saying they will, but that is the kind of thing that would start to get people interested in switching.

Re: Only in the garden of Eden.

Big technology companies aren't interested in present day, they have to look 5 even 10 years in advance. If they looked what users wanted now then by the time they delivered it to the users it would be 2 years out of date.

Re: I'm sure this is true, in some shape or form. But that's not the point.

"I'm sure they could commission a custom x86 CPU with some Apple IP if they wanted one with a very consolidated design-even if Intel said no AMD would find it hard to turn down the money right now."

Tricky - Apple doesn't have an x86 design license, so they couldn't do their own without spending a *lot* of money on buying one.

"From their end, I suppose switching to custom ARM chips might offer savings-but it leaves them dependent on their own innovation fighting against the whole PC industry, just like in the PPC days, with a risk of it all running out of steam."

They already have custom ARM chips - in iPhone and iPad. Ironically Apple bought PASemi which gave them a competitive PowerPC design (it would have been great in laptops), but they killed that straight away. They might ride the ARM-in-server wave that's building, but might not go in a desktop friendly direction.

"Adding uncertainty when they're on the verge of stealing loads of disgruntled Windows users would be...audacious."

Disgruntled Windows users? You can still buy Windows 7, and that's been very popular. You're right though - Windows 8 is certainly unfamiliar and disconcerting compared to the old and stuffy OS X and Windows 7.

"I think this a headline planted to make Intel jump into line, not a serious threat."

Perhaps, but I'm not sure that Intel can 'jump into line'. X86 is exceedingly difficult to make fast and low power, the instruction set is just way too crummy for that. And if Intel fixed that by changing the instruction set it then wouldn't be an x86 any more... Their latest effort has completely bombed because it's still way too power hungry. They don't need any encouragement from Apple to know that they're missing out big time.

But Intel (and AMD) are still very good if you have a mains supply and don't mind about the power consumption too much; their chips have a *lot* of compute power which makes for nice games, etc, and I can't see ARM bothering trying to compete with that any time soon.

Re: I'm sure this is true, in some shape or form. But that's not the point.

custom x86 CPU Unless they buy AMD, no chance in the world

Re: Could be interesting

tagged memory? sounds like the paging, sandboxing and virtualization that are already being done. sometimes even in hardware. There's definitely space for a lot more interesting stuff to go on in this department, but it's going to take an exponential increase in memory density and especially throughput to make it happen.

Re: Just buy AMD already

That would definitely be a cheaper way of going independent. After all AMD has no expensive and cumbersome fab to weigh Apple down, their designs are pretty good really, and would come complete with a very competitive GPU too.

I suspect that it means that they don't want to remain x64 compatible, which begs the question, what? Presumably that means ARM, and that in itself raises certain questions.

Apple make the bulk of their cash from iPhone and iPad, not from iMac, PowerMac or the laptops. The iPhone/iPad are closed ecosystems, you have to go through the iStore for literally everything you ever want to put on them, and that earns Apple a bundle of cash too. Even the botched move to iMaps is related to ongoing income from previously sold products, and is currently a failed attempt to stop Google making money of the back of iPhones.

In contrast the iMac/etc. are open in the sense that you can install any software you like. However Apple don't make a penny from it afterwards. Every one uses Google (money going to Google), buys MS Office (money going to Microsoft), buys Photoshop (money going to Adobe), buys games (money going to EA or whoever), etc. In fact the only post sales income Apple gets through their desktop / laptop lines is iTunes sales and the occasional OS upgrade. I think that's a powerful indicator of their intent.

If this rumour / news is true I think it means that Apple want to kill of Mac OS X as we know it, and turn their non-mobile products into closed ecosystems too, and a change to the ARM architecture (or indeed any other architecture, even PowerPC) is a good excuse for doing just that.

I 'upgraded' from a a Powermac G5 quad to a MacPro 3,1 and it was by no means a massive step forward in performance. In some ways, it was quite a bit slower. The PPG 970fx was an inefficient chip, but a powerful one. The major difference between the two machines was the enormous drop in power consumption.

Re: OSX on ARM. Why not?

"Apple have switched architectures more than once before. It's something they seem quite comfortable doing."

Yes, they pulled that off with 68000->PowerPC->Intel, but every time the amount of compute power available increased. That eased the pain because it meant that emulators had good performance. An Intel->ARM transition wouldn't be able to do that. ARM cores are not fast, and emulation would have to run on the ARM core itself and that'd be a very slow dog indeed.

"Switching to ARM would give them a lot of potential benefits in addition to owning the silicon."

Others have already pointed out that buying AMD would give them the same control, but allow them to stick with x86.

"How many ARM processors can you fit into the same silicon/price bracket as an Intel processor?"

A lot. An ARM core has something like 48,000 transistors, whereas an x86 core contains many millions. Intel's problem is that they also need millions more transistors (cache, instruction decoding, etc) to make the x86 core fast.

"I'd be very surprised if a similarly priced array of ARMs wasn't as powerful as the Intel equivalent, and they'd probably use less power too."

It maybe as powerful in aggregate, but single thread performance would still be bad. That matters - most software is still single threaded, so all your apps would run slow. That might not be an issue for something that's not compute-heavy (a text editor). For other stuff like video decoding one of ARM's hardware accelerators would do the job, just like a mobile. However, something like Photoshop could be a real dog.

ARM or Apple might seriously pep up the ARM core to reduce the disparity in single thread grunt, and it would be more power efficient than Intel. However it sounds like a lot of work, and it be probably be cheaper to just go back to PowerPC; Freescale haven't been standing still with PowerPC, and if it comes to that Apple bought PA Semi a few years back - they had a very competitive PowerPC design at the time.

Re: No need to kill OS X

Maybe, but its programs don't necessarily run on anything, especially if they're binary-only. That's machine code, and most machine code is meant for only ONE type of machine architecture. It's a rare beast that can do more than one in hardware.

Re: Nail in coffin for business

" it proves once and for all that Apple don't care about business continuity or stability."

I thought OSX server (RIP) already did that?

PowerPC?

Intel has managed to mitigate the weakness of the X86 instruction set by having the best fabs, and not making chips for anyone else. They even sold off their ARM based products to Marvell, to prove their faith in X86 to their sceptical customers. But if losing Apple was a done deal, and Apple don't sell bare chips, maybe Intel could make ARM, or even PowerPC SOC's for Apple.

However much you may dislike Apple's control freakery, it's what enabled them twice before to switch CPU architectures. Something both Microsoft and Intel have aspired to, but failed to gain traction with. (remember Windows NT for PowerPC? iAPX432, Xscale, Itanium?)

The nub of the matter is: what could Apple persuade Intel to do for them, given that X86 is not remotely the mainstream any more, and iGizmos must inexorably move towards zero build cost over the decades.

Re: I wonder if it can be done.

IBM had the stomach. What they didn't have was the customers to buy PowerPC desktop/laptop systems in the kind of volume needed to justify the capital investment to keep on par with Intel. They need far more business than Apple could supply and nobody else with the needed cash was inclined to jump in. IBM had already strangled their own effort in its crib years earlier.

Without the right numbers no amount of intestinal fortitude would make it a business worth pursuing.

Re: Emulation

Have you seen an x86-on-ARM emulator or DynaRec core in action? You're talking an efficiency core trying to emulate a performance core. In the last two instances, Apple transitioned to clearly-superior hardware which smoothed the transition. Here? ARM isn't necessarily superior to Intel: just different. That handicaps emulation: just as trying to emulate a high-speed PowerPC or MIPS CPU even on today's Intel CPUs is not a walk in the park.

Re: Rosetta

You have to take into account the memory model into "multi-core era", ARM has significantly weaker memory model compared to x86 (x64 is a bit stronger even). Concurrent code cannot be JITed just like that single proper memory fences have to be issued. x86 never reorder load w/ load or write w/ writes unlike ARM.

Re: Peltier cooling

Unfortunately all that does is makes the heatsink even hotter as the heat is actively dragged from the CPU. Not much good for thin and light designs, because while the CPU might be running at negative-degrees celcius, the heatsink would be rapidly melting a hole in the chassis.

Now heatpipes on the other hand, they can help. Even if only to spread the radiators around the inside of the chassis to distribute and diffuse that nasty hotspot.

Re: Count me out...

"(stupid scrolling, full screen rows of icons for no real reason, locked down default behavior for external apps, no real save as, removal of hard drive icon on desktop to 'stop people getting confused' etc etc"

None of this is true. Trust me, Mountain Lion is what you want. Snow Leopard is no longer adequate when you have used ML. There's a little bit of culture shock, but things like iOS style notifications are actually excellent, and all the restrictions you cite are untrue.

many Windows applications have parts written in x86

not really. It's more about Microsoft not very keen on inter-architecture machine code translation, but very keen on maintaining its advantage in the number of available applications already compiled to x86/x64.

In other words: users will stay with Windows as long as they can continue using applications they accrued over the time, some of them expensive (e.g. work-related) or large collections of cheaper ones (e.g. games). Since these are already purchased, the vendors have little incentive for rebuilding the binary for new platform. In case if Windows was to jump the platform it would upset vendors and customers by forcing one to rebuild/port and the other to purchase applications again. The effect would be Windows losing its advantage , i.e. number of available applications. You can observe that by watching how the situation with Windows RT develops.

As for Microsoft's "increasing reliance on .NET" , that has stopped with Longhorn failure. It was the last time Microsoft tried to do system programming in .NET and it was a write off. Windows 8 is move back to native , e.g. WinRT is not .NET but native code with COM-like wrappers for .NET . Surprisingly to some, that is true even on Windows RT .