Apple A11 Bionic performance

[MarkizSchnitzel]

Has it not been said multiple times that cores in the Apple A SOC-s are huge, and that is the main (not only, but main) reason for monstrous performance?
The conclusion was that good part of the performance difference could be crossed with a "wider" core and more transistors, so it's simple economics that are holding Intel/Qualcomm behind?

I have no interest in Apple ecosystem, and know next to nothing about SOC design, but I am impressed with what Apple is doing, so I am curious, as it will also impact everyone's computing habits, weather one uses Apple products or not.

Are they simply that good at SOC achitecture?
Is vertical integration as important as I would assume?
Or is it the economics that are the biggest advantage that Apple has?

First one would cost tons of money for competitors to beat.
Second one.. Google seems to be going this route of vertical integration, slowly, with Pixel devices?
Third one.. if Apple creates new market segments, economies of scale would allow others to come closer with their own SOCs.

If you had to assign numerical values to these 3, how would you do it?

Is there something else I am missing?

 

[beginner99]

While lightroom will be using a heavily optimised GPU accelerated path on the iPad.

That and fixed-function hardware for encoding.

 

[Denly]

This is where Microsoft have really dropped the ball. They had a golden opportunity to do this with Windows 8.1/10 and have failed to really do it. Contniuum is from all reports limited and buggy.

That was their chance to get back into the mobile game and they blew it.

the issue with continuum is not bug, it is the app. It only support UWP, so company's x86 software don't work on it. If you only need office and email then yet it work pretty good.

 

[Av9114]

I can only say it's true. I have an Ipad 4 (retina, a6x) which has score in gb4 820 single and 1360 multi and an honor 7 phone which has 780 single and 2200 multi core score in GB4

so the honor 7 should be faster, but it is not

web browsing on the a6x is much faster than on the kirin 935

IMO it is not only about apple CPUs, but also with the direct optimization of iOS for the hardware

Safari, at least mobile Safari's javascript engine seems to improve at an impressive rate year to year. When anandtech posts new iPhone reviews they don't re-run the javascript tests on the old devices with the new OS, if you do have one of the older devices in their charts and re-run the tests you see that there are big gains on the software side as well.

FYI, I just ran JetStream on my laptop and Safari seems to have a pretty big edge over Chrome (~50%) here as well.

 

[Av9114]

Nevermind. User error...

 

[thunng8]

That and fixed-function hardware for encoding.

Why it is faster is interesting but in the end result is the most important metric.

Lightroom is faster on an iPad Pro than the MacBook Pro

 

[beginner99]

Why it is faster is interesting but in the end result is the most important metric.

Lightroom is faster on an iPad Pro than the MacBook Pro

Why is for sure important if we are arguing about the CPU / SOC. If we argue about the platform which includes much more, then yes speed is the only metric that matters.

Since we are in the first scenario, the why matters a lot. And simply said comparing the iPad Pro to a laptop is like comparing a console to a PC. Console is 1 set of hardware which you can optimize to the metal and hence you will always be a lot more efficient and can beat faster hardware running less optimized code.

In this case I suspect many stuff on macOS isn't yet GPU-accelerated or the still bad intel iGPU is just much slower than A11 which would be absolutely no surprise.

 

[StrangerGuy]

Has it not been said multiple times that cores in the Apple A SOC-s are huge, and that is the main (not only, but main) reason for monstrous performance?
The conclusion was that good part of the performance difference could be crossed with a "wider" core and more transistors, so it's simple economics that are holding Intel/Qualcomm behind?

Saying Apple has much better performance because of a wider design/larger core is 100% pure sour grapes talk from the non-Apple camp, when the Ax cores are still tiny compared to the rest of the SoC they are in. You think anybody else wouldn't want 2x ST performance for a few more mm^2 of silicon real estate per core if they had the choice?

 

[Av9114]

Has it not been said multiple times that cores in the Apple A SOC-s are huge, and that is the main (not only, but main) reason for monstrous performance?
The conclusion was that good part of the performance difference could be crossed with a "wider" core and more transistors, so it's simple economics that are holding Intel/Qualcomm behind?

While I think it's true that the Ax series CPU cores are generally larger than stock ARM designs I think solely attributing the performance to that is a vast oversimplification. Certainly Intel would be willing to double their CPU core sizes if they could get 50% more ST performance at the same power. The problem with wide cores is that it's generally hard to keep them fed, and any idle parts are just wasting power. I think this is part of the reason that Intel and now AMD have SMT/Hyperthreading or whatever. It's actually pretty interesting that Apple doesn't as far as I know. So I think it's probably more accurate to say that Ax performance is largely because Apple has found a way to make a big core and keep it fed without using obscene amounts of power.

Are they simply that good at SOC achitecture?
Is vertical integration as important as I would assume?
Or is it the economics that are the biggest advantage that Apple has?

First one would cost tons of money for competitors to beat.
Second one.. Google seems to be going this route of vertical integration, slowly, with Pixel devices?
Third one.. if Apple creates new market segments, economies of scale would allow others to come closer with their own SOCs.

If you had to assign numerical values to these 3, how would you do it?

Is there something else I am missing?

At this point it seems virtually certain that Apple's CPU architects are the best in the world. One might argue for Intel, but it seems pretty tough at this point. I think the real question is how much of it is "skill/talent" and how much of it is other factors, which I think is what you're getting at above. I would think that CPU performance would get the least amount of benefit from vertical integration, after all as far as I know they're running the same instruction set as all other ARM vendors and the compiler is open source. The big advantage is that they get to tailor their designs to a relatively narrow target. These designs only need to work in iPhone/iPad size devices and therefore they may be able to make different design decisions than a company like Intel which tries to use more or less the same basic core design all the way from tablet to server (ARM/Qualcomm/Samsung don't have this excuse though). Until proven otherwise it's probably best to assume that you can't just pump 200W into an A11 and push it to 5Ghz, for example. There are lots of other small examples of things like this too, and I have no idea how much they matter. Economics matter some, but the biggest factor there is probably that product development is so expensive that it limits the potential players i.e it's not very likely that a little startup comes out of nowhere to dominate. I doubt that Apple spends so much on their designs that Intel/Qualcomm/Samsung would be unwilling to match it for similar results.

 

[Arachnotronic]

Saying Apple has much better performance because of a wider design/larger core is 100% pure sour grapes talk from the non-Apple camp, when the Ax cores are still tiny compared to the rest of the SoC they are in. You think anybody else wouldn't want 2x ST performance for a few more mm^2 of silicon real estate per core if they had the choice?

The fact that Apple can build such a wide machine, keep it in a phone-level power envelope, and hit the same clocks that the other, much narrower designs can hit is nothing short of mind-blowing.

 

[JoeRambo]

he big advantage is that they get to tailor their designs to a relatively narrow target.

And it's not about HW targets as well. They are in full control of OS and can make sweeping changes with ease. For example it is pretty much impossible for Intel and AMD to change basic page size, it is stuck at 4KB as it is in DNA of x86 and need to be supported real well.
While Apple moved to 16KB pages and suddenly their same entry TLB coverage grew by factor of 4, probably enabling some other benefits like cache hit/miss latency reduction. They might pay the price for larger pages somewhere else, but they can mitigate it with full OS control more than other ARM guys.

Currently advantage of AXX CPUs is beyond ridiculous, it's really sad one can't buy Android phone with it.

 

[Arachnotronic]

And it's not about HW targets as well. They are in full control of OS and can make sweeping changes with ease. For example it is pretty much impossible for Intel and AMD to change basic page size, it is stuck at 4KB as it is in DNA of x86 and need to be supported real well.
While Apple moved to 16KB pages and suddenly their same entry TLB coverage grew by factor of 4, probably enabling some other benefits like cache hit/miss latency reduction. They might pay the price for larger pages somewhere else, but they can mitigate it with full OS control more than other ARM guys.

Currently advantage of AXX CPUs is beyond ridiculous, it's really sad one can't buy Android phone with it.

They can also do tricks like removing 32-bit support altogether, something that will be very tough in the wild, wild west of Android or, even moreso, Windows.

 

[moinmoin]

Why it is faster is interesting but in the end result is the most important metric.

Indeed. In the end everything, may it be GPU acceleration, fixed function hardware, AVX 512 etc. all needs software support to make good use of. Apple's ability to vertically integrate OS level software and all the development toolsets with hardware and vice versa is (unfortunately) unrivaled. This also affording them the removal of 32bit support both at hardware and OS level is spot on.

 

[MarkizSchnitzel]

I learned something in this thread.

But gotta ask a slightly off topic question, because it's been raised that end result is what matters most.

With that huge advantage that is the AXX, and vertical integration, and full control, why does my OP3 with Nougat still "feel" faster and more fluid than IP7+ running iOS11? Where did they make a mistake?
I realize that most tasks will be faster on the iphone (applying filters, compressing and whatnot), but why does the UX/UI not benefit as much from that? A colleagues of mine has a new ip7+ and side by side, there is noticeable stuttering on the iphone. And it's been the same on ip6 and iphone6s that he had before this. So it's not something new.

 

[Verne Arase]

Gawd ... I hope Macs don't go ARM.

I need my x86 hypervisor.

 

[Verne Arase]

One of the advantages Apple has is that it can build specialized blocks to suit its needs and because it controls the OS, it can develop APIs to expose that functionality, quickly move first-party applications to take advantage of that hardware via those APIs, and then app developers -- looking to not be left behind by other app developers -- will quickly adopt these technologies to give their apps every benefit that they can get in a very crowded app marketplace.

Another thing is the adoption by users.

iOS 11 is currently in use by 34% of iOS users - a slower rate of adoption than iOS 10, but not terrible after two weeks. Between iOS 11 and iOS 10, you can address 94% of all iOS users.

Still, I think that iOS 11 adoption dwarfs the adoption of any single version of Android.

What's the current adoption rate for Oreo? .2%?

If you're a developer, do you target Oreo under the assumption that those are the users who might pay for your efforts, or do you target Kitkat 4.4 at 14.5%, Lolipop 5.0 at 6.7%, Lolipop 5.1 at 21%, Marshmallow 6.0 at 32%, or Nougat 7.0 at 15.8%? What API do you target, and how much of an audience will you enjoy?

Android is a long convoy of boats, held together by bubblegum and baling wire. The more you pull on one end, the more boats snap off and get left in their own stand-alone islands.

Is it worthwhile to try to make money with ARCore? How many devices have the oomph to run it reasonably well, and how far back can you assume the runtimes will be available?

 

[Verne Arase]

With that huge advantage that is the AXX, and vertical integration, and full control, why does my OP3 with Nougat still "feel" faster and more fluid than IP7+ running iOS11? Where did they make a mistake?
I realize that most tasks will be faster on the iphone (applying filters, compressing and whatnot), but why does the UX/UI not benefit as much from that? A colleagues of mine has a new ip7+ and side by side, there is noticeable stuttering on the iphone. And it's been the same on ip6 and iphone6s that he had before this. So it's not something new.

Betcha he's got something he's let run in the background that's eating up memory/CPU cycles.

Apple's approach to the kernel is one of the more interesting I've seen in a while.

They don't shoot for spec benchmarks like other unix variants - they shoot for real-time user responsiveness (reduction of the beachball), even if they have to drill down into the kernel to get it.

One big advantage of the iPhone is you can take the backup of your old phone and simply restore it on your new phone. And you can keep doing it ad nauseam. The same with macOS - you can keep just applying one OS on top of another.

Unfortunately that means if you have a problem, it can follow you for a long time.

Tell him to wipe his phone to factory specs and put apps back on one at a time. Alternately, he can check what's using the most CPU time and start wiping them out one by one. You can build a goodly amount of cruft over time, and every so often it makes sense to start over.

 

[StrangerGuy]

Another thing is the adoption by users.

iOS 11 is currently in use by 34% of iOS users - a slower rate of adoption than iOS 10, but not terrible after two weeks. Between iOS 11 and iOS 10, you can address 94% of all iOS users.

Still, I think that iOS 11 adoption dwarfs the adoption of any single version of Android.

What's the current adoption rate for Oreo? .2%?

If you're a developer, do you target Oreo under the assumption that those are the users who might pay for your efforts, or do you target Kitkat 4.4 at 14.5%, Lolipop 5.0 at 6.7%, Lolipop 5.1 at 21%, Marshmallow 6.0 at 32%, or Nougat 7.0 at 15.8%? What API do you target, and how much of an audience will you enjoy?

Android is a long convoy of boats, held together by bubblegum and baling wire. The more you pull on one end, the more boats snap off and get left in their own stand-alone islands.

Is it worthwhile to try to make money with ARCore? How many devices have the oomph to run it reasonably well, and how far back can you assume the runtimes will be available?

The Android ecosystem has too much conflict of interests within itself. Which is made worse when the OS developer cares about their own datamining dollars first and everything else a distant second.

If it weren't for Apple HW's distancing itself increasingly away from Android year after year Google wouldn't have been frightened enough to release the Pixel to grab back mindshare, which ironically IMO it's nothing but an iPhone wannabe.

 

[Emacs25]

Betcha he's got something he's let run in the background that's eating up memory/CPU cycles.

Apple's approach to the kernel is one of the more interesting I've seen in a while.

They don't shoot for spec benchmarks like other unix variants - they shoot for real-time user responsiveness (reduction of the beachball), even if they have to drill down into the kernel to get it.

Unfortunately it is not the case anymore. iPhone 6s was smooth, iPhone 7 and now iPhone 8 are not smooth, they both have noticeable random frame drops on animations.

This is something phone specific though, may be 3D touch, because my iPad 10.5 Pro is much better in this regard: it can drop frames, but it only happens at the beginning (and, in addition, calling any 60Hz device "buttery smooth" looks like a joke after 120Hz of the iPad).

 

[moinmoin]

Regarding refresh rates and frame drops it's curious that Apple with their vertical integration aren't using variable refresh rates on the screens for their iDevices yet.

 

[NTMBK]

Regarding refresh rates and frame drops it's curious that Apple with their vertical integration aren't using variable refresh rates on the screens for their iDevices yet.

I thought the the ProMotion on the iPad used VRR?

(Side note: It's really hard to get information on that feature, because Googling "Apple ProMotion" gets you loads of results about offer codes and student discounts. Great job, marketing department *facepalm* )

 

[IntelUser2000]

Well, personally I despise Apple, so it ticks me off every time I see a post saying how great their CPUs are..

Disliking it doesn't change the truth however.

I think this is part of the reason that Intel and now AMD have SMT/Hyperthreading or whatever. It's actually pretty interesting that Apple doesn't as far as I know. So I think it's probably more accurate to say that Ax performance is largely because Apple has found a way to make a big core and keep it fed without using obscene amounts of power.

I think its pretty certain a team as capable as they one they have realizes problems with CPU utilization. Regardless, they have a even better solution than SMT to improve multi-core performance.

The 4x Mistral low power cores work like extra logical threads in an SMT-enabled CPU. The fact that they made such a setup work is amazing. I assume because it uses entirely separate cores it would likely even avoid performance degradation due to resource contention as with SMT when the cores are fully utilized. Plus, they can even use those cores to lower power use during light-usage scenarios above and beyond(or is it below ) what Intel/AMD chips can do.

At this point it seems virtually certain that Apple's CPU architects are the best in the world. One might argue for Intel, but it seems pretty tough at this point. I think the real question is how much of it is "skill/talent" and how much of it is other factors, which I think is what you're getting at above.

Skill/Money/Tool are just a few things that can influence success. I think there's something just as important. You can have the smartest engineers, with best tools and funding, but they won't do much with it if they are in a depressed state, or no confidence in the future or for the place they are working. Apple is pretty much THE darling of the current tech industry consisting of mobile devices.

 

[moinmoin]

I thought the the ProMotion on the iPad used VRR?

(Side note: It's really hard to get information on that feature, because Googling "Apple ProMotion" gets you loads of results about offer codes and student discounts. Great job, marketing department *facepalm* )

Good call, it does indeed (yeah, crappy name, and I apparently completely forgot it as part of the whole Apple Pencil push). As of now it appears to be limited to the high end iPads and the latest iPhones whereas I'd think it's of good use across the board especially for relatively low powered devices (Apple Watch maybe). I guess I'm just surprised it took them so long, and now it's just a matter of the tech trickling down the product stack over time.

 

[DrMrLordX]

Well, personally I despise Apple, so it ticks me off every time I see a post saying how great their CPUs are..

You'd better get used to it. Depending on how ambitious they are, Apple could expand their line of CPUs into other markets. Especially when they see those fat margins just waiting for them.

 

[Verne Arase]

The problem with wide cores is that it's generally hard to keep them fed, and any idle parts are just wasting power. I think this is part of the reason that Intel and now AMD have SMT/Hyperthreading or whatever. It's actually pretty interesting that Apple doesn't as far as I know. So I think it's probably more accurate to say that Ax performance is largely because Apple has found a way to make a big core and keep it fed without using obscene amounts of power.

Keeping your cores fed has always been the system tuner's nightmare ... even back in the IBM mainframe days when they were engines (not cores) and they ran on bipolar water-cooled TCMs.

It's the reason that multitasking was developed - to keep those expensive CPUs busy by suspending a task while some slow operation was dispatched to the I/O subsystem (like disk/tape I/O) by passing the torch to the next task on the ready queue.

At first, I/O completion interrupts were used by the dispatcher, but as CPU and I/O subsystems got faster they added instructions to test pending completion interrupts like TPI so they could avoid the overhead of running the interrupt.

Multiprocessing added an additional nightmare - the same code could be running the same instruction on reentrant code simultaneously, so serialization of previously trivial things could become important - and the old trick of masking out interrupts was laid by the wayside since that only worked on a per processor basis.

System semaphores like enqueue allowed serialization of resources like dataset control, but as multiprocessing progressed they needed serialization of more ephemeral resources so lock loops were created with instructions like CS for things like the storage lock for memory allocation coherency within an address space.

Who'd have guessed back in the day that we'd have portable phones which had multiprocessing operating systems, and that these pocket devices would have the same problems with process flow as all that big iron ... and that some OS developers of those devices would fall into the same pitfalls as some of those early systems of yore?