Apple mulling transition away from Intel chips for Macs

12346»

Comments

  • Reply 101 of 119
    ascii wrote: »
    I'm not sure you really want to go up against Chipzilla. They will figure out *some* way to make x64 as power efficient as ARM, just you watch, they'll pull something out of the hat.

    They'd have done it by now. If they were as good as you say.
  • Reply 102 of 119
    ecsecs Posts: 307member


    Shocking that everybody is worried about the possible Intel->ARM transition, which actually isn't of any concern because if/when that happens, performance of both architectures will be similar.


     


    However, the article _does_ have a worrying affirmation that everybody is neglecting: The wish of Bob Mansfield: Convergence of iOS and OSX. This is exactly what Tim Cook is criticizing about Windows 8: You shouldn't use the same paradigm for devices that are used in a different way. We already had enough of it with Lion. It seems Mansfield wants to go that road, and that road has just one meaning: Forget the Mac, say hello to the iPad Pro.


     


    Now, this _is_ worrying. But the Intel->ARM transition is totally harmless to the Mac, I don't care about it.

  • Reply 103 of 119
    vaelianvaelian Posts: 446member
    jragosta wrote: »
    It will also lead to a massive decrease in performance - which would not be acceptable to most people.

    While that's true on current hardware, it doesn't have to be in future hardware. As far as design is concerned, ARM is a lot easier to implement than x86; the only thing currently keeping x86 in the lead is Intel's vastly superior experience and competence in processor design, but given another company with comparable experience I have no doubt that ARM can completely destroy x86, both due to being a much simpler design with a much smaller instruction set and without backward compatibility concerns and because it's easier to optimize compilers for RISC. Also, from what I'm told, Apple is betting on massively parallel implementations with 32, 64, 128 core SoCs for Macs. Last I heard, a lot was going on inside Apple to make existing code as parallel as possible with this in mind.

    jragosta wrote: »
    There is, of course, always the possibility of using AMD chips, though.

    AMD themselves are betting on ARM64 for future servers.

    kolchak wrote: »
    On the one hand, there's the problem that this would eliminate Bootcamp and virtualization options. That's not a small loss for many users. On the other hand, this would also eliminate the possibility of Hackintoshes, but those don't take a big bite out of Apple's sales anyway.

    Microsoft themselves are changing the Windows paradigm.
  • Reply 104 of 119
    backtomacbacktomac Posts: 4,579member

    Quote:

    Originally Posted by Marvin View Post

    .... This 'Swift' architecture is early days but a pretty good effort:

    http://www.anandtech.com/show/6330/the-iphone-5-review/4


     


    I surprised you posted an Anandtech link to the iPhone 5 review to support your belief that ARM could be an acceptable chip for Macs.


     


    Surely you noticed in the performance benchmarks how competitive the Atom cpu in the Motorola handset is? And by competitive I mean faster in several benchmarks.


     


    People shouldn't be wondering when ARM will be power Macs. They should be wondering when Intel cpus will be in iOS devices. Later in 2013 Intel will be rolling out a major Atom redesign. It will be fabbed at the 22nm process node. I predict that chip will be the fastest chip for mobile phones when it is released. I think it is going to easily surpass the best available ARM chip at that point in time. 

  • Reply 105 of 119
    jnjnjnjnjnjn Posts: 588member

    Quote:

    Originally Posted by alandail View Post


     


    in most cases it would just be a recompile.


     


    Certainly there are important exceptions, things like photoshop and office.  But for the most part apps already are built to support MacOS X on Intel and iOS on ARM.  And developers who know what they're doing don't write code that would care which CPU it ran on.


     


    And Xcode already knows how to build binaries that support multiple CPU architectures.  Announce at WWDC and ship in the fall and 90% of shipping apps would already be updated by the time machines shipped.  I'm sure people like Adobe and Microsoft would take longer, but that wouldn't matter too much because if Apple did this, the first machines would most likely be more affordable MacBook Airs. which already don't target photoshop users.


     


    as for the cpu power that keeps being mentioned.  Certainly Apple wouldn't do something like this without a more powerful CPU than the iPad 4 has.  But a 4 core 64 but ARM CPU should be plenty powerful for a more affordable entry level machine.  


     


    Building their own CPUs has proven to be quite important for Apple in the iOS world.  It wouldn't' be at all surprising if they wanted to bring that to the MacOS world. 



     


    Excellent comment.


     


    The porting problems of Adobe and Microsoft had to do with differences in APS's (or, to be precise, the conversion from Carbon to Cocoa); they effectively had to reimplement the applications.


    A transition from x86 code to ARM will be as simple as switching Xcode to ARM and waiting a few minutes, if Apple converts the full OS X API set to ARM (and they probably did that already). 


    Certainly, CPU power is important, but most CPU intensive tasks are offloaded by iOS and OS X to the GPU's. This is done via OpenGL and OpenCL.


    So GPU performance is very important and current generation PowerVR cores have desktop performance.


    A7 next year could be a 64bit A57 based PowerVR rogue SOC and will be off desktop performance, probably using a fifth of the energy of a comparable Intel processor. I expect the MacBook Airs to make the transition to ARM first.


     


    J. 

  • Reply 106 of 119
    jragostajragosta Posts: 10,473member
    You mean by turning Apple into another commodity Wintel PC assembler?

    Nice theory. Except for the fact that they've never been "on intel's roadmap" for any of their iOS devices, and yet these devices drive Apple's earnings. If the post-PC vision ever becomes a reality, how important will being "on intel's roadmap" be for Apple? How important will intel be in the post-PC world?

    So Apple is just another commodity Wintel PC assembler? I guess that's why everyone had to copy the MBA - and Intel had to offer $300 M in subsidies for Apple's competitors. I guess it's why there are so many 27" AIO computers out there that look just like the iMac. :rolleyes:
    ksec wrote: »
    The potential savings from switching to ARM are billions per year. Calculating at 10 Million Macs and a saving of $100 per devices. The actual figures could varies since we dont know what sort of discount Intel are giving Apple.

    So you don't have any idea what you're talking about, but that won't stop you from making things up......

    But have it your way. Apple saves $100 per device on 10 M Macs per year - or $1 B per year. But they lose 2 M Mac sales at $1500 each - or $3 B. See? I can make numbers up, too. I suspect mine are closer to the truth, though.

    If Apple wants to save money on processors, there are better ways to do it. Use slightly older Intel chips - which would still outperform ARM. Use AMD chips.
    vaelian wrote: »
    While that's true on current hardware, it doesn't have to be in future hardware.

    Yeah. ARM is going to advance mightily in coming years while Intel won't improve their chips at all. :no:
  • Reply 107 of 119
    jnjnjnjnjnjn Posts: 588member

    Quote:

    Originally Posted by Vaelian View Post





    While that's true on current hardware, it doesn't have to be in future hardware. As far as design is concerned, ARM is a lot easier to implement than x86; the only thing currently keeping x86 in the lead is Intel's vastly superior experience and competence in processor design, but given another company with comparable experience I have no doubt that ARM can completely destroy x86, both due to being a much simpler design with a much smaller instruction set and without backward compatibility concerns and because it's easier to optimize compilers for RISC. Also, from what I'm told, Apple is betting on massively parallel implementations with 32, 64, 128 core SoCs for Macs. Last I heard, a lot was going on inside Apple to make existing code as parallel as possible with this in mind.

    AMD themselves are betting on ARM64 for future servers.

    Microsoft themselves are changing the Windows paradigm.


     


    Your absolutely right.


    Apple uses Grand Central Dispatch (GCD) and a small C extension (blocks) to make writing parallel code a lot easier.


    So using lots of cores on a SoC is a very good idea to enhance performance.


     


    J. 

  • Reply 108 of 119
    backtomacbacktomac Posts: 4,579member

    Quote:

    Originally Posted by Vaelian View Post





    While that's true on current hardware, it doesn't have to be in future hardware. As far as design is concerned, ARM is a lot easier to implement than x86; the only thing currently keeping x86 in the lead is Intel's vastly superior experience and competence in processor design, but given another company with comparable experience I have no doubt that ARM can completely destroy x86, both due to being a much simpler design with a much smaller instruction set and without backward compatibility concerns and because it's easier to optimize compilers for RISC. Also, from what I'm told, Apple is betting on massively parallel implementations with 32, 64, 128 core SoCs for Macs. Last I heard, a lot was going on inside Apple to make existing code as parallel as possible with this in mind.

     


    If adding cores were all that is necessary to make the fastest computers, we'd all be using gpus to do our tasks. Its not that easy. Some tasks cannot be made parallel in code. Single core performance is still important. Just ask AMD. And I wouldn't be looking at AMD to get a glimpse of the future of computing. Just because they are throwing in with ARM doesn't make it a winner over x86 for server applications. They've been wrong before.

  • Reply 109 of 119
    MarvinMarvin Posts: 15,310moderator
    backtomac wrote: »
    Marvin wrote: »
    .... This 'Swift' architecture is early days but a pretty good effort:
    http://www.anandtech.com/show/6330/the-iphone-5-review/4

    I surprised you posted an Anandtech link to the iPhone 5 review to support your belief that ARM could be an acceptable chip for Macs.

    Surely you noticed in the performance benchmarks how competitive the Atom cpu in the Motorola handset is? And by competitive I mean faster in several benchmarks.

    People shouldn't be wondering when ARM will be power Macs. They should be wondering when Intel cpus will be in iOS devices. Later in 2013 Intel will be rolling out a major Atom redesign. It will be fabbed at the 22nm process node. I predict that chip will be the fastest chip for mobile phones when it is released. I think it is going to easily surpass the best available ARM chip at that point in time. 

    Apple using Intel's low power chips in iPads and iPhones is a possibility and would be the easier route towards Mac and iOS binary compatibility. As far as the benchmarks go, the Intel Atom is dual 1.3GHz that dynamically clocks to 2GHz. The A6 is dual 1GHz that dynamically clocks to 1.3GHz and it's not clear what the power draw is of each. I'd bet that performance per watt is better in Apple's CPU.

    No comparison right now will make ARM seem like an obvious choice for switching but just like with the PPC to Intel switch, it's all down to the roadmap. I wouldn't look forward to updating every Mac app again and wouldn't care about doing that for the iOS apps so moving from ARM to Intel would be the less painful option but if ARM chips can be 2x faster for a lower price, I think it would be worthwhile.

    The emulation headache won't be anywhere near as bad as before either. This happened in 2005 when computers were less than 1/10th the speed of their modern equivalents. The lower-end dual 1.8GHz G5 ($2000, 2004) scored 0.38 in Cinebench 11. The current middle Mac Mini ($800) scores about 6.2 so a factor of 16x in 8 years (double every 2 years and those models aren't equivalent in price). A lot of people back then were struggling with single processor G4s, which score less than half the G5.

    Emulating Windows on a G4 using VirtualPC in 2004/2005 would be nothing like emulating Windows on a Core i7 in 2012 and we're not talking about the transition happening in 2012.

    If it turns out to be worth doing, I just don't think it's something to be too concerned about.
  • Reply 110 of 119
    backtomacbacktomac Posts: 4,579member

    Quote:

    Originally Posted by Marvin View Post





    .... As far as the benchmarks go, the Intel Atom is dual 1.3GHz that dynamically clocks to 2GHz. The A6 is dual 1GHz that dynamically clocks to 1.3GHz and it's not clear what the power draw is of each. I'd bet that performance per watt is better in Apple's CPU.


    The Atom in the Mototrola Razor I is a single core chip with hyperthreading. Link.


     


    I think Intel are going to leap frog ARM in performance when the next version of Atom. It is supposedly going to be dual core and an Out of Order design. Heck if it is single core with hyperthreading and OoO I think it will surpass ARM. 


     


    I think people here are assuming a larger architectural advantage for RISC over CISC chips than actually exists these days. The over head that CISC chips have compared to RISC is relatively small these days. IIRC its only about 10%. Performance now will largely be dependant on how small the chips are fabbed. I don't see Intel relinquishing their advantage here anytime soon.

  • Reply 111 of 119
    tt92618tt92618 Posts: 444member

    Quote:

    Originally Posted by Tallest Skil View Post




    Really, it sort of is. Macs don't sell because of Boot Camp.



     


    I think it is a bigger deal than you suggest, especially in the enterprise.  If my company couldn't run Windows in virtualization or boot camp without decent performance, there wouldn't be a single mac on anyone's desk or in anyone's laptop bag.  It is absolutely that big of a deal.

  • Reply 112 of 119

    Quote:

    Originally Posted by Realistic View Post




    ARM doesn't run iOS otherwise there would already be Macs running on ARM chips. Show us any links saying, showing or even claiming that ARM chips can run OS X.



    I believe you mistyped your claim that ARM doesn't run iOS.


     


    There have been numerous rumors of MacBook Air chassis running OS X on ARM. It would be hard to believe that Apple was running parallel PowerPC/X86 development of OS X, then forked OS X to ARM for iOS, but did not pursue full OS X on ARM. That would be a lot less work than the parallel PowerPC/X86 projects.


     


    And here's another interesting angle on ARM/X86... http://www.eetimes.com/electronics-news/4397620/Russian-software-runs-x86-code-on-ARM

  • Reply 113 of 119
    jragostajragosta Posts: 10,473member
    bsimpsen wrote: »
    I believe you mistyped your claim that ARM doesn't run iOS.

    There have been numerous rumors of MacBook Air chassis running OS X on ARM. It would be hard to believe that Apple was running parallel PowerPC/X86 development of OS X, then forked OS X to ARM for iOS, but did not pursue full OS X on ARM. That would be a lot less work than the parallel PowerPC/X86 projects.

    And here's another interesting angle on ARM/X86... http://www.eetimes.com/electronics-news/4397620/Russian-software-runs-x86-code-on-ARM

    Yes, but note:
    "Elbrus Technologies has developed emulation software that currently delivers 40 percent of native ARM performance. "

    40% is pretty good for emulation, but that means that ARM (which is already far slower than Intel solutions) would have its performance reduced by 60% - or more (if they're using favorable results).


    "The company believes it could reach 80 percent native ARM performance or greater by the end of 2014."

    Lots of companies have made similar claims, but no one has yet produced an emulation solution with that level of performance on any platform. Even their claimed 40% is higher than anyone else has achieved. I guess I'll believe it when I see it.
  • Reply 114 of 119
    jragostajragosta Posts: 10,473member
    Marvin wrote: »
    Apple using Intel's low power chips in iPads and iPhones is a possibility and would be the easier route towards Mac and iOS binary compatibility. As far as the benchmarks go, the Intel Atom is dual 1.3GHz that dynamically clocks to 2GHz. The A6 is dual 1GHz that dynamically clocks to 1.3GHz and it's not clear what the power draw is of each. I'd bet that performance per watt is better in Apple's CPU.
    No comparison right now will make ARM seem like an obvious choice for switching but just like with the PPC to Intel switch, it's all down to the roadmap. I wouldn't look forward to updating every Mac app again and wouldn't care about doing that for the iOS apps so moving from ARM to Intel would be the less painful option but if ARM chips can be 2x faster for a lower price, I think it would be worthwhile.
    The emulation headache won't be anywhere near as bad as before either. This happened in 2005 when computers were less than 1/10th the speed of their modern equivalents. The lower-end dual 1.8GHz G5 ($2000, 2004) scored 0.38 in Cinebench 11. The current middle Mac Mini ($800) scores about 6.2 so a factor of 16x in 8 years (double every 2 years and those models aren't equivalent in price). A lot of people back then were struggling with single processor G4s, which score less than half the G5.
    Emulating Windows on a G4 using VirtualPC in 2004/2005 would be nothing like emulating Windows on a Core i7 in 2012 and we're not talking about the transition happening in 2012.
    If it turns out to be worth doing, I just don't think it's something to be too concerned about.

    I think you're grossly underestimating the difficulties.

    Regardless of what you're starting with, you lose something like 2/3 to 3/4 of your performance when you emulate an instruction set. Since ARM is so much slower than Intel to start, the problem would be even greater.

    Yes, it's true that today's computers are far faster than 15 year old computers. And, yes, one could easily emulate a 15 year old computer with today's computers - and it would seem fast. But no one is buying 15 year old computers. They're buying modern computers and x86 emulated on ARM would be horrendously slow compared to anything on the market.
  • Reply 115 of 119
    MarvinMarvin Posts: 15,310moderator
    jragosta wrote:
    Regardless of what you're starting with, you lose something like 2/3 to 3/4 of your performance when you emulate an instruction set. Since ARM is so much slower than Intel to start, the problem would be even greater.

    The switch would be under the assumption that at some point in the future, ARM would be faster than Intel. If that never comes to pass then there's no point but the performance improvements we've seen so far suggest that it's a possibility.
    jragosta wrote:
    Yes, it's true that today's computers are far faster than 15 year old computers. And, yes, one could easily emulate a 15 year old computer with today's computers - and it would seem fast. But no one is buying 15 year old computers. They're buying modern computers and x86 emulated on ARM would be horrendously slow compared to anything on the market.

    If you assume that you run at 40% performance, that's equivalent to a 3-year old computer.

    We know that Intel has been doubling performance fairly consistently every 2 years. This is shown in benchmarks for the past decade or more. Mobile ARM processors have been doubling in performance every year since 2007. If those rates of improvement continue, the smartphone class ARM CPU will outperform the laptop class Intel CPU in 5 years. TSMC has already made a dual 3.1GHz ARM chip. Apple could probably clock their A6 at 3.1GHz with a power draw of 15W tops and it would work as a laptop CPU and right now would be comparable in performance to the Macbook Air. If they double the cores, it moves into Macbook Pro territory.

    There's no sense in doing anything until the performance is significantly higher but that could be just 3 years away. If the performance is double an equivalent Intel chip, running at 40% via emulation = 2 x 0.4 = 80% native.

    Usage cases for Bootcamp should be considered too. For gaming it's going to be a bad deal. For things like accounting and probably even modelling apps like 3DS Max, it'll be fine. If you link up with say V-Ray on the Mac side, you can run at native speeds. Windows used to run like the following on a G4:




    For average Windows usage, it should run pretty smoothly on a sufficiently fast ARM chip especially now that we have loads of RAM and SSDs. Emulation mainly won't be suitable for resource-intensive tasks but most of those can be done on the Mac side.

    Also, bear in mind that Intel is slowing down on CPU improvements in favour of the IGP. They've dropped year-on-year CPU bumps to 10-15%. Graphics is another thing to consider for a switch to ARM but they could easily use a GPU from AMD/NVidia given that the power consumption of the CPU would be less. Instead of the Mini having a 45W quad-i7 and an Intel IGP, it could have a 15W quad ARM and a 30W AMD/NVidia dedicated GPU. While they could just use an AMD CPU, if the ARM performance per watt is higher than Intel, it will certainly be higher than AMD.
  • Reply 116 of 119
    jragostajragosta Posts: 10,473member
    Marvin wrote: »
    The switch would be under the assumption that at some point in the future, ARM would be faster than Intel. If that never comes to pass then there's no point but the performance improvements we've seen so far suggest that it's a possibility.
    If you assume that you run at 40% performance, that's equivalent to a 3-year old computer.
    We know that Intel has been doubling performance fairly consistently every 2 years. This is shown in benchmarks for the past decade or more. Mobile ARM processors have been doubling in performance every year since 2007. If those rates of improvement continue, the smartphone class ARM CPU will outperform the laptop class Intel CPU in 5 years. TSMC has already made a dual 3.1GHz ARM chip. Apple could probably clock their A6 at 3.1GHz with a power draw of 15W tops and it would work as a laptop CPU and right now would be comparable in performance to the Macbook Air. If they double the cores, it moves into Macbook Pro territory.
    There's no sense in doing anything until the performance is significantly higher but that could be just 3 years away. If the performance is double an equivalent Intel chip, running at 40% via emulation = 2 x 0.4 = 80% native.

    How about a little critical thinking.

    Do you really think that ARM will have twice the performance of Intel in 3 years? That's so implausible that it hardly merits mention.
  • Reply 117 of 119
    MarvinMarvin Posts: 15,310moderator
    jragosta wrote: »
    How about a little critical thinking.

    Do you really think that ARM will have twice the performance of Intel in 3 years? That's so implausible that it hardly merits mention.

    They are about even just now comparing like for like e.g comparing Medfield to the A6 and not an i7 to the A6.

    If Intel focuses on improving their IGP and CPU performance goes up 15% each year, in 3 years, they will be at 1.15 ^ 3 = 1.52x faster. If ARM improves just 50% each year, in 3 years they will be at 1.5 ^ 3 = 3.38x faster. 3.38 / 1.52 = ARM is 2.22x faster than Intel.
  • Reply 118 of 119

    Quote:

    Originally Posted by Cash907 View Post



    Not going to happen. The Mac platform exploded after adopting intel chips


     


    Yes, because Intel's chips featured far better performance-per-watt than IBM was delivering with PPC.  That allowed Apple to expand dramatically in the then fast-growing laptop segment, where they now dominate the industry in terms of profit.


     


    Today, ARM's chips feature far better performance-per-watt in the smartphone and tablet space than anything Intel has been able to bring to market.  If ARM continues to release ever more capable chips with similar performance-per-watt, they'll soon be the only viable option in the ultralight notebook market.  And at that point, given that iOS is Apple's largest business by far anyhow, it'll certainly make sense for them to switch their entire product line over to customized ARM chips.


     


    Even if ARM's raw performance never matches that of Intel's best x86 chips, Apple could still offer comparable end-user performance on their larger laptops and desktop machines by larding them with high-end GPUs and other support chips, and by deploying multiple ARM CPUs on their high power machines.  Since they won't be paying the Intel tax anymore and won't need to maintain two separate code bases, they can buy a lot of extra CPUs with that kind of money.  Two ARM CPUs with just more than half the performance of an Intel CPU will likely provide a better end user experience than that single Intel CPU, and could end up costing Apple somewhat less as well.


     


    It would also give Apple total control over its roadmap.  That's worth billions in and of itself.

  • Reply 119 of 119

    Quote:

    Originally Posted by ecs View Post


    However, the article _does_ have a worrying affirmation that everybody is neglecting: The wish of Bob Mansfield: Convergence of iOS and OSX. This is exactly what Tim Cook is criticizing about Windows 8: You shouldn't use the same paradigm for devices that are used in a different way. We already had enough of it with Lion. It seems Mansfield wants to go that road, and that road has just one meaning: Forget the Mac, say hello to the iPad Pro.



     


    We have no idea what the convergence of iOS and OS X entails.  It would be possible to merge the two without merging the actual UI at all.  The OS could detect how you're interfacing with the device - voice, touch, or keyboard and mouse - and adjust the UI accordingly.  So the 'desktop' UI would look nothing like the touch UI.


     


    This is how Microsoft should have approached Windows 8.  But MS has never understood user interfaces, and it shouldn't be a surprise to anyone that they've botched their attempt to integrate their mobile and desktop UIs.  Apple likely won't make the same stupid mistake.

Sign In or Register to comment.