Highly suspect benchmarks stoke rumors of Apple-designed ARM chips for Mac

24

Comments

  • Reply 21 of 65
    mcdavemcdave Posts: 1,170member
    The GPU scores don’t make sense for an iMac that’s a fair downgrade.  I was thinking 6/8 performance + 4 efficiency cores + 12-core GPU but that’s pretty weak scalability on the CPU more like a boosted 4+4, no HBM2?  This looks more like a MacBook Pro/ low end iMac.
    Thought most serious software vendors are supporting or plan to support ARM, I would have thought x64 emulation would be pre-requisite.
    watto_cobra
  • Reply 22 of 65
    melgrossmelgross Posts: 31,803member
    jkichline said:
    Seems to me that if you’re going to transition to ARM, you need enough horsepower to handle x86 emulation for apps not recompiled to support ARM. I suppose this would be trivial to recompile existing apps using an updated version of Xcode, or to compile iOS apps to Mac soon which already using ARM instructions.
    Despite what many people think, most apps are not a recompile away. Yes, small, simpler apps may be. But think about all of the demo’s we’ve seen over the years from software developers who, on stage, said; ...and we did this in one weekend, it was so easy! And then the actual app didn’t arrive for 6 months. Because it’s NOT so easy. Recompiling for a different chip family is never easy. 

    The instruction set is different. Some instructions aren’t even similar. X86 is Ciscier, while ARM is Riscier. Moving from one to the other is not simple for bigger apps. So big apps such as Office, and Photoshop, and Final Cut will have to run under emulation for some time, at half speed. We’ve seen this several times now, so don’t be surprised.

    putting these into a Macbook, which uses a weak CPU could work, because this would be a lot more powerful, so that emulation would be fine. Big apps likely wouldn’t suffer much.
    muthuk_vanalingamwatto_cobra
  • Reply 23 of 65
    nsummy2nsummy2 Posts: 11member
    bugsnw said:
    It's exciting to imagine a world where desktops and laptops follow iPhones on the hardware upgrade cycle. Even if the upgrades for non-phones were every 2 years, consistently, that would be better than some of the huge delays that occur now. Modular designs where components could receive updates for a certain time would be helpful in extending the usefulness of our hardware. I can think of one particular instance where Intel really put the screws to Apple: my MacBook was maxed out at 16GB ram and really needed 32 for opening up browsers with dozens of tabs. From what I understand, Intel's chip limited Apple on that front.
    You need to pull your head out of the sand. Everything you mentioned is apple's doing, not Intel's.  If you want hardware that is regularly updated, modular, and allows expansion, buy a Dell, hp, Lenovo, etc.  And if you like os x that much you can run it in a vm :)
    GeorgeBMaclkruppwilliamlondon
  • Reply 24 of 65
    nsummy2nsummy2 Posts: 11member
    These obviously aren't benchmarks for a new apple laptop but I guess good for page views and clicks. That said I can't wait to see how this eventually shakes out.  I just can't imagine desktops and pro grade laptops running arm processors.  Aside from the cpu, how do you handle the graphics? Apple isn't going to be able to create anything that beats Nvidia or AMD.   The only way I see this happening is if Apple is creating a new class of ultrabooks that join the ranks of what microsoft is trying to accomplish with windows on arm.

    On top of that, what happens if software performance is terrible on the arm chips?  It would be catastrophic for Apple if an Adobe product worked twice as fast on x86.  Like I said, it will be interesting to see how all of these factors are addressed. And I'm just scraping the surface.
  • Reply 25 of 65
    evn616evn616 Posts: 26member
    I welcome this for the Mac Pro. I would upgrade my i9 hex core machine for a powerful ARM variant. Intel is a “has been”. I much prefer AMD in the x86 world over Intel. I think providing a hardware accelerated solution for x86 emulation on an ARM platform would be appropriate to bridge the gap until most relevant software has been recompiled.
    watto_cobra
  • Reply 26 of 65
    thttht Posts: 3,243member
    melgross said:
    tht said:
    Information provided by an anonymous uploader claims the 10-core version is clocked at 3.4GHz, while supposed Geekbench screenshots show the 12-core chip, referenced as "[email protected]," running at 3.19GHz. The two application processors achieved respective single-core scores of 7335 and 6912, and multi-core scores of 20580 and 24240.

    The multicore scores don’t make sense if the single core scores are ~7000 GB4 points. So fake imo. 

    For the 10-core version to achieve something like this, it would be what, 2 big cores and 8 small cores? For the 12 core model, maybe 3 big cores and 9 small cores?

    This would perhaps be fitting for a laptop. For desktop, no reason to spare the cores. Just have 6, 8, 12, 16, etc, especially if perf/watt is better than Intel Skylake. 
    Ok, I was going to post substantially the same thing. Those scores make no sense. The way Apple is going with the A+ series chips, there are equal numbers of performance and efficiency cores (not BIG/LITTLE, that’s an ARM naming convention for cores that work differently than Apple’s). So that could mean 5/5, or 6/6. If so, then the multi core scores should be higher. Since the more cores, the smaller percentage of contribution, ten of those cores should give about 6 times the performance. Twelve should give about 7 times.

    unless Apple is doing something odd. Actually, why should there even be as many efficiency cores as performance cores, since notebooks aren’t as limited in power and heat as iPads are. Besides, with increased clock speed, even efficiency cores should have enough oomph so that not a lot would be needed, and I would imagine that increased performance would come from improved IP as well.
    If it is a real, the simplest explanation is that this is just test hardware for the 2019 iPad Pro, in either a 2:8 or 3:9 performance:efficiency core setup, or for every performance core, there are either 4 or 3 efficiency cores. It’s a battery powered device of some kind as it has a battery score. Piling up the efficiency cores could be a design decision to support new iPad Pro features, like multi-app, multi-document and long term background app capabilities.

    If it is an ARM laptop, I have no doubts Apple wants it be lighter, faster, cooler and run longer than an Intel equivalent, so this type of setup is fine. Just, not sure why so many efficiency cores would be needed.
    watto_cobra
  • Reply 27 of 65
    melgross said:
    jkichline said:
    Seems to me that if you’re going to transition to ARM, you need enough horsepower to handle x86 emulation for apps not recompiled to support ARM. I suppose this would be trivial to recompile existing apps using an updated version of Xcode, or to compile iOS apps to Mac soon which already using ARM instructions.
    Despite what many people think, most apps are not a recompile away. Yes, small, simpler apps may be. But think about all of the demo’s we’ve seen over the years from software developers who, on stage, said; ...and we did this in one weekend, it was so easy! And then the actual app didn’t arrive for 6 months. Because it’s NOT so easy. Recompiling for a different chip family is never easy. 

    The instruction set is different. Some instructions aren’t even similar. X86 is Ciscier, while ARM is Riscier. Moving from one to the other is not simple for bigger apps. So big apps such as Office, and Photoshop, and Final Cut will have to run under emulation for some time, at half speed. We’ve seen this several times now, so don’t be surprised.

    putting these into a Macbook, which uses a weak CPU could work, because this would be a lot more powerful, so that emulation would be fine. Big apps likely wouldn’t suffer much.
    The size and complexity of the app is 100% irrelevant for porting between chips if the code is 100% portable. 

    it’s only ever relevant that porting is difficult if they’ve coded low-level bit-twiddling and special SIMD math operations in CPU-specific instructions, because often compilers don’t emit optimized instruction streams from code to do that: not all C/C++ compilers are created equal.

    The most common practice for applications include first writing and testing algorithms in whatever higher-level language being used (where “higher-level” is higher than assembly) and then you profile the application, and only if the higher-level language implementation doesn’t provide high enough performance, would you then rewrite the time-critical inner loops in the local CPU assembly language using special instructions the compiler doesn’t know how to translate the higher-level coded algorithm into the optimized assembly that a compiler does for most things.  The practical reality is these days with speculative execution and out-of-order-execution, humans are going to produce less optimized code than a modern compiler does.

    The higher-level code is #ifdef’ed out (C/C++/Objective-C/C++) and the optimized assembly code is #def’ed in, if a developer has it: this is normal practice.

    Very little code in any application of any size is going to be written in hand-written assembly: doing more than that is non-sensical from an ROI standpoint.  How muc is likely to be in assembly? A small fraction of 1%.  1% would be a HUGE amount of assembly in an application >10,000 lines of code.
    thtchiaGeorgeBMacPickUrPoisonwatto_cobra
  • Reply 28 of 65
    thttht Posts: 3,243member
    mcdave said:
    The GPU scores don’t make sense for an iMac that’s a fair downgrade.  I was thinking 6/8 performance + 4 efficiency cores + 12-core GPU but that’s pretty weak scalability on the CPU more like a boosted 4+4, no HBM2?  This looks more like a MacBook Pro/ low end iMac.
    Thought most serious software vendors are supporting or plan to support ARM, I would have thought x64 emulation would be pre-requisite.

    The GPU score looks to be for an on-chip GPU (like in all of Apple’s ARM SoCs). If so, that’s good for 13” laptops or base model desktops. It’s about equivalent to the Radeon Pro 555X, maybe the 560X, in the MBP15. The leaker reported a battery benchmark, and since when does a desktop run on batteries? So if real, it’s not a desktop machine.

    They’ll have JIT style (like Rosetta in the past) x86 to ARM instruction translation and probably an x86 Mac emulater too. Though I imagine hosting iOS apps could go a long ways in address consumer needs for apps, and those folks wouldn’t even care for Mac apps that aren’t updated.
    watto_cobra
  • Reply 29 of 65
    mcdavemcdave Posts: 1,170member
    melgross said:
    jkichline said:
    Seems to me that if you’re going to transition to ARM, you need enough horsepower to handle x86 emulation for apps not recompiled to support ARM. I suppose this would be trivial to recompile existing apps using an updated version of Xcode, or to compile iOS apps to Mac soon which already using ARM instructions.
    Despite what many people think, most apps are not a recompile away. Yes, small, simpler apps may be. But think about all of the demo’s we’ve seen over the years from software developers who, on stage, said; ...and we did this in one weekend, it was so easy! And then the actual app didn’t arrive for 6 months. Because it’s NOT so easy. Recompiling for a different chip family is never easy. 

    The instruction set is different. Some instructions aren’t even similar. X86 is Ciscier, while ARM is Riscier. Moving from one to the other is not simple for bigger apps. So big apps such as Office, and Photoshop, and Final Cut will have to run under emulation for some time, at half speed. We’ve seen this several times now, so don’t be surprised.

    putting these into a Macbook, which uses a weak CPU could work, because this would be a lot more powerful, so that emulation would be fine. Big apps likely wouldn’t suffer much.
    Actually, most apps are easy to recompile that’s the point of an HLL.  The key dependencies are; the portability of any 3rd-party libraries/frameworks, how much ISA-specific code is used and how much of the performance is down to optimisation which tightens the code relationship with the target architecture.  The latter usually takes the time.
    watto_cobra
  • Reply 30 of 65
    mcdavemcdave Posts: 1,170member

    nsummy2 said:
    These obviously aren't benchmarks for a new apple laptop but I guess good for page views and clicks. That said I can't wait to see how this eventually shakes out.  I just can't imagine desktops and pro grade laptops running arm processors.  Aside from the cpu, how do you handle the graphics? Apple isn't going to be able to create anything that beats Nvidia or AMD.   The only way I see this happening is if Apple is creating a new class of ultrabooks that join the ranks of what microsoft is trying to accomplish with windows on arm.

    On top of that, what happens if software performance is terrible on the arm chips?  It would be catastrophic for Apple if an Adobe product worked twice as fast on x86.  Like I said, it will be interesting to see how all of these factors are addressed. And I'm just scraping the surface.
    Then you’re very poorly informed.  They look exactly like benchmarks for a new MacBook Pro/Mac Mini.  Apple’s GPU tech is within striking distance of AMDs genuine mobile processors (Nvidia can’t be used off the power cord) and way ahead in performance per watt. Those Metal scores align with a 12-core GPU_Family_4 part and the architecture scales to 16 (12 has been used in the A10X).
    The A12X shows how far Apple are ahead of Intel U-series in CPU performance and a larger power budget/core-count gets them across the line for HQ/K.  I’m not thinking a 3+9 config though, that makes little sense to me.  6xBig+4xLittle+12xGPU should fit in a 15W power budget (MBP13) but deliver MBP15 performance.
    You’re right about x64 ISA compatibility though.  Re-compilation is a mid/long-range strategy so they will need to emulate in the short term but please don’t assume Qualcomm are the benchmark in any way, their mobile SoCs are 2-3 years behind Apple.
    williamlondoncornchipwatto_cobra
  • Reply 31 of 65
    viclauyycviclauyyc Posts: 360member
    An unsubstantiated, and highly suspect, "leak" on Friday claims to reveal benchmarks from a pair of desktop-class ARM processors supposedly designed by Apple, offering what could be the first look at A-series silicon destined for Mac.

    ARM


    Outlined in a post to Slashleaks on Friday are supposed Geekbench benchmarks for ARM big.LITTLE chips with 10- and 12-core architectures.

    Information provided by an anonymous uploader claims the 10-core version is clocked at 3.4GHz, while supposed Geekbench screenshots show the 12-core chip, referenced as "[email protected]," running at 3.19GHz. The two application processors achieved respective single-core scores of 7335 and 6912, and multi-core scores of 20580 and 24240.

    What device the alleged ARM chips are powering is unknown, but the performance of each falls in line with desktop class hardware. Both processors beat single-core benchmarks set by Apple's 2017 27-inch Retina 5K iMac with 4.2GHz Intel Core i7-7700K, while the purported 12-core ARM chip's mutli-core score bests the 2018 15-inch MacBook Pro with 2.9GHz Intel Core i9-8950HK.

    It is unclear if the benchmarks shared today are legitimate, but it should be noted that the uploader joined SlashleaksAppleInsider was unable to verify the leak and does not vouch for its veracity.

    Apple has long been rumored to transition Mac away from Intel to a bespoke ARM architecture, but hard evidence of the development process has been elusive.

    Last October, analyst Ming-Chi Kuo in a note to investors predicted an ARM-based mac to arrive in 2020 or 2021. More recently, Intel officials in February told Axios they expect the tech giant to make the switch "as soon as next year."
    I am sure there are many ARM base laptop/desktop class machines running in Apple HQ. But if Apple really stop using Intel/X86 cpu, that means a big chunk of corporate/power user will stop Apple products for good. The ability to run windows, Linux and other x86 base OS on a quality Mac machine is why these companies willing to pay extra bucks for it, not MacOS.

    Will Apple willing to make the switch just for being “I am special” like the good old days? I doubt it. After all, it is not like Intel has bad relationships with apple or not able to improve like Motorola/IBM.
    macplusplus
  • Reply 32 of 65
    sflocal said:
    My only hope is that there will still be some kind of compatibility with x86 instruction set.  I know I'm part of a minority group, but I do have all my Macs (three of them) running Windows for certain development tools I use.
    I agree. I know that for Linux, we will probably have it soon after losing intel CPU, but losing Windows compatibility will hurt many people.
    edited March 16 macplusplusAbalos65
  • Reply 33 of 65
    MisterKitMisterKit Posts: 269member
    The ‘leak’ is exciting if not verified because this is what Apple had better been doing over the last 5-10 years. 
    watto_cobra
  • Reply 34 of 65
    19831983 Posts: 1,185member
    Interesting! I hope these benchmarks are legit. I’m pleasantly surprised to see that a single core ARM processor can already match or beat a desktop Intel equivalent...if of course, this is the real deal.
    watto_cobra
  • Reply 35 of 65
    franco borgo said:

    I agree. I know that for Linux, we will probably have it soon after losing intel CPU, but losing Windows compatibility will hurt many people.   

    Yes, this will be important for certain users. Although Windows 10 can already be run on ARM with x86 emulation for programs not recompiled, maybe it would be possible to bootcamp with this version of Windows 10. 



    GeorgeBMaccornchipwatto_cobra
  • Reply 36 of 65
    GeorgeBMacGeorgeBMac Posts: 4,967member
    Well, this is the only explanation I could think of why an iMac refresh is taking so long.

    Although I can dream these benchmarks are for a new Apple TV.
    AppleTV already has the A8. What about an Apple TV is lacking in performance? I doubt AppleTV would ever get something like this. It may get a refresh to A11 or later, but that will be to make it easier on developers who use hand-rolled Metal code. (A11 architecture brought significant changes/advances to how Metal code is written).
    I didn't know that it could drive Fortnite, Call of Duty or GTA on a 4K TV.
    williamlondonwatto_cobra
  • Reply 37 of 65
    benji888benji888 Posts: 114member
    sflocal said:
    I firmly believe that Apple has had a MacOS/ARM machine in development for several years.  Just like it did with the IBM->Intel, when the time is right and the performance is at least on par with Intel's offerings, they will introduce that machine.  It only makes sense that it is the next logical step in Apple's eventual divorce from major chip suppliers like Qualcomm and Intel.  Intel has shot itself in the foot way too many times and proven to be a real headache for Apple in having a constant introduction of newer x86 chips.
    ...

    I was going to say the same thing...we found out after the shift from PowerPC, (Motorola/IBM), to x86/Intel that Apple had been compiling a version of Mac OS X to run on x86 for years before the actual transition...so, whether this benchmark is real or fake, Apple is surely compiling a version of macOS for its own arm based chips...the difference here is this time Apple is producing their own chips, so they could be prototyping the chips and the OS, whereas before it was just the OS. Plus, with iOS and macOS being built off the same core, it should be an easier transition.

    However, I wouldn’t expect to see this in the first iteration of the new Mac Pro, it’s coming this year, macOS on arm is still a couple years or so away.
    cornchipwatto_cobra
  • Reply 38 of 65
    GeorgeBMacGeorgeBMac Posts: 4,967member
    melgross said:
    jkichline said:
    Seems to me that if you’re going to transition to ARM, you need enough horsepower to handle x86 emulation for apps not recompiled to support ARM. I suppose this would be trivial to recompile existing apps using an updated version of Xcode, or to compile iOS apps to Mac soon which already using ARM instructions.
    Despite what many people think, most apps are not a recompile away. Yes, small, simpler apps may be. But think about all of the demo’s we’ve seen over the years from software developers who, on stage, said; ...and we did this in one weekend, it was so easy! And then the actual app didn’t arrive for 6 months. Because it’s NOT so easy. Recompiling for a different chip family is never easy. 

    The instruction set is different. Some instructions aren’t even similar. X86 is Ciscier, while ARM is Riscier. Moving from one to the other is not simple for bigger apps. So big apps such as Office, and Photoshop, and Final Cut will have to run under emulation for some time, at half speed. We’ve seen this several times now, so don’t be surprised.

    putting these into a Macbook, which uses a weak CPU could work, because this would be a lot more powerful, so that emulation would be fine. Big apps likely wouldn’t suffer much.
    While that is likely true, I suspect the majority of Mac users would simply shrug --- because they don't use them.   Increasingly the only ones using MS-Office are those who need it to be compatible with their office.   For the rest, Apple's free look-alike products will do just fine.
    watto_cobra
  • Reply 39 of 65
    GeorgeBMacGeorgeBMac Posts: 4,967member
    nsummy2 said:
    These obviously aren't benchmarks for a new apple laptop but I guess good for page views and clicks. That said I can't wait to see how this eventually shakes out.  I just can't imagine desktops and pro grade laptops running arm processors.  Aside from the cpu, how do you handle the graphics? Apple isn't going to be able to create anything that beats Nvidia or AMD.   The only way I see this happening is if Apple is creating a new class of ultrabooks that join the ranks of what microsoft is trying to accomplish with windows on arm.

    On top of that, what happens if software performance is terrible on the arm chips?  It would be catastrophic for Apple if an Adobe product worked twice as fast on x86.  Like I said, it will be interesting to see how all of these factors are addressed. And I'm just scraping the surface.
    Why would it be catastrophic?   Few people use any Adobe product except Reader and there are plenty of substitutes for that.

    Apple is about meeting people's needs, not performance specs.
    cornchipwatto_cobra
  • Reply 40 of 65
    I would also doubt that the first ARM-based Macs will be Mac Pros or even Macbook Pros. Typical light users — the same ones who were the target audience for the Macbook, who value portability above all else and who use primarily a few core apps like Safari, Mail, Messages, possibly MS Office — are a much better fit. Apple can transition their own apps first and already have 90% of the functionality such users need without having to wait for even the first third party developer. Beyond this the most central failing of Intel's chips has been on the low-power end, and without doubt Apple's AX chip could restart stalled innovation in super-portable laptops. It would be ironic if the A-series introduction to Macs provides a long-missing substantive reason to the current arbitrary split between pro and non-pro Macs, but presence of Intel chips distinguishing the pro lines would certainly make the distinctions clear.
    GeorgeBMaccornchipwatto_cobra
Sign In or Register to comment.