For several years, Apple has been steadily designing more and more of the chips powering its iPhones, iPads, Macs and Apple Watches. This creates a better user experience and helps trump rivals. Recently the company got a fresh incentive to go all-in on silicon: revelations that microprocessors with components designed by Intel Corp., Arm Holdings Plc and Advanced Micro Devices, Inc. are vulnerable to hacking.
[…]
That original “system-on-a-chip†has since been succeeded by increasingly powerful processors. Today, Apple packs its devices with custom components that process artificial intelligence tasks, track your steps, power game graphics, secure Face ID or Touch ID data, run the Apple Watch, pair AirPods to your phone and help make Macs work the way they do. The result: a chip powerhouse that could one day threaten the dominance of Qualcomm Inc. and even, eventually, Intel.
Apple’s chip business really puts the company in a unique position. No other phone or PC maker can rely on such a powerful chip division, with the exception of Samsung, but Samsung’s own ARM chips are nowhere near as powerful as Apple’s. Assuming Apple manages to turn their chip prowess into real-world advantages for users, it’ll be hard for competitors to catch up.
Benchmarks are basically meaningless because they are incredibly easy to manipulate. It is mostly a matter of overclocking the cores and killing background processes to get a huge (temporary) increase in performance. In the real world this causes overheating, very poor battery life and a largely unusable device.
In realistic scenarios Apple devices perform about the same as the other flagships.Some manufacturers (e.g. HTC) use very conservative settings which make their devices appear slower.
Edited 2018-01-31 02:35 UTC
Also the benchmarks normally performed on phones and tablets are not comparable between platforms at all. As someone working on embedded browsers, I can tell you that if I run an mobile browser emulated locally, it scores higher in performance than the native desktop browser.
I hope people don’t think that Apple processors are more secure. We know that their software is far from it, so to expect them to be better at something that is infinitely more difficult is foolish.
I suspect that Samsung and ARM does almost all of the design and engineering work on the ‘Apple’ chips.
Apple bought PA Semi years ago and they had the best RISC cpu engineers in the world.
But hey, bashing Apple is so cool…
Do you think Apple just feeds the blueprints into a machine and a CPU magically pops out?
It is quite reasonable to assume to that the vast bulk of the engineering process is converting a design into a working product – just like in every other engineering field.
The cynic in me thinks that a lot of Apple’s engineering efforts are primarily for PR purposes and that most of the complex engineering work is done by suppliers. Tim Cook has already said that it would be impossible to build Apple devices in the US due to a lack of engineering skills.
I’m sure there are half a dozen companies that could make chips just as good. But then Apple couldn’t boast about the (alleged) superiority of their own designs.
Edited 2018-01-31 08:49 UTC
as mentioned above the work is done by PA Semi who are one of the best chip designers out there, apple purchased them and they are the engineers / designers of the A series chips.
Designing a chip is just a very small part of the overall production process. In this case Apple are modifying an existing ARM design.
The vast bulk of the engineering work is actually getting the design from the drawing board into mass production. That is where Samsung’s expertise comes into play.
Right, but wait… A11 is made by TSMC.
The A11 is an ARM reference design modified by Apple.
There is a huge amount of work to convert a mask to a finished chip. TSMC engineers and technicians would be providing many thousands of hours of work to refine the design and deliver a finished product.
The A5 was an ARM reference design modified by Apple. Your only about 6 years off… Every design since then has been fully custom.
The A11 is an ARM reference design modified by Apple.
Ahaha. Which reference design?
You pulled this out of thin air. But the reality is different.
Apple has architecture license like Qualcomm and others.
TSMC engineers and technicians would be providing many thousands of hours of work to refine the design and deliver a finished product.
TSMC engineers spent thousands of hours to create component library. And they provide it to customer before he can start the physical implementation (or even start the microarchitecture design).
TSMC can’t refine the design. Factory do not have source code and they are not performing the synthesis.
Edited 2018-02-02 02:53 UTC
Here is the problem with your fantasy view of how this process works…
Samsung does NOT fabricate the A11 – TSMC does. Samsung literally has nothing whatsoever to do with it. At all. Not a f*cking thing. Not sure how Samsung would be able to help them with their “expertise” when they don’t even participate in building the chip…
They also had nothing to do with the A10 either. Samsung hasn’t been involved with the fabrication of Apple CPUs since the A9 (iPhone 6s), and even then that CPU was fabricated by BOTH Samsung and TSMC. Samsung has NEVER designed a CPU in any Apple device. Not a single one. The CPUs before the A6 where used cookie cutter ARM cores – Samsung just fabbed them (like they still occasionally do)… There has never been a Samsung desgined CPU in an Apple device. Hell, Samsung rarely uses their own designs in THEIR phones (they almost always use Qualcomm designs in their US phones)
Your argument is that the vast majority of the work is getting from the drawing board to mass production. This is absolutely true, but the fab has virtually nothing to do with that part of the process. There are many fabs out there than can do mass production, but getting from the drawing board to the point that you can submit a mask for production is entirely up to the design team. They target the particular process used by the Fab to a degree, but with modern designs this is become less and less of an issue (hence why 2 entirely different companies could produce the A9 with little to no differences between them).
Look, I don’t much care about you deriding Apple’s engineering abilities – have fun in your delusion… However, if your going to do so at least try to sound like you have a clue what you are talking about.
tldr; Apple designs some good CPUs. They have one of the best ARM CPU design teams on earth. This is simply just fact, that you chose to give other’s credit for it is your own personal prerogative – it doesn’t reflect reality at all. You are just making yourself sound stupid by trying to tear them down without a shred of facts or knowledge to back any of it up…
Edited 2018-02-01 03:14 UTC
The educated engineer with knowledge of how the actual process works in my thinks you are completely full of shit and talking authoritatively about things you know nothing at all about.. Just stop already.
Actually, on this one point, he’s right and you’re DEAD wrong. You don’t just insert a blueprint and MAGICALLY out pops a chip of any kind. You’ve clearly never designed a chip and had it fabricated, or done the fabrication side yourself. You’ve clearly never done a PC board for that matter. You don’t just pop in a net schematic and magically out pops a PC board. Go work with some autorouter software for a while; make a few 6 layer boards, then come back and remark about how that “magic” that makes board and chips pop out is or isn’t a major amount of the work.
EDIT: Had a great example that shows the issue clearly.
ARCHITECT: Here’s the blueprints for the 200 story high rise. The major work is done. Now magically pop out that 200 story building!
CONTRACTOR: Yeah – it doesn’t work like that.
Edited 2018-02-01 17:47 UTC
Of course not… Does your sarcasm detector not work at all?
The equipment required to get from a mask to a chip is complex as hell and costs millions and millions of dollars. There is no magic involved, it is a highly sophisticated process that takes a tremendous amount of investment and engineering.
But my point is this: once you have a mask, magic or not, the process of getting from that mask to a chip does not involve a bunch of engineers doing design work on the chip – that part of the process is done already. They are doing fabrication work, work that is essentially the same no matter what mask is currently being processed. The engineering expertise of the fab went into designing the process, not the chip…
The process for getting from a mask to a working chip is expensive, complicated, and requires a whole lot of work. It does not, however, have anything AT ALL to do with CPU design.
Edited 2018-02-02 01:12 UTC
Doesn’t change that they have bought all the building blocks and only need to work on optimizing key components and the general design. So you could say that most of the chip was done by others, but you could also argue that Apple is done most if not all the important stuff themselves since they are able to outperform others using the same building-blocks.
So what are these “off-the-shelf” building blocks Apple is using for their CPU cores? Logic gates?
The factory is providing that for synthesis.
I find this whole discussion fascinating – people who literally know nothing about how this process works grasping to cobble together anything that might explain how Apple could possibly design a CPU…
The explanation is rather mundane – they actually know how to design CPUs. How that is hard for anyone to believe when its public knowledge they bought an establish 150 person design team 10 years ago (one of the best in the world mind you) I don’t quite understand.
There are people who are not comfortable with the idea that Apple has CPU engineers at all.
So they believe in a magic “ARM reference design” system there Jonny Ive just switching a couple of levers to get the leading performance CPU. Then Samsung is doing the rest of job.
But they can’t explain why other CPU teams can’t switch the same levers and put the “basic blocks” together in the same combination to get equally fast CPU =)
Hi,
More importantly, we know that all of Apple’s CPUs are vulnerable to both Meltdown and Spectre (see: https://www.techarp.com/guides/complete-meltdown-spectre-cpu-list/3/ ).
– Brendan
Edited 2018-01-31 16:38 UTC
Nowhere it states that “all” are vulnerable to “both”.
A4, A5 are ARM designs (Cortex-A8, Cortex-A9) and they are not affected by Meltdown.
https://developer.arm.com/support/security-update
I don’t know yet which Apple-designed cores are vulnerable by meltdown, but is definitely not all.
Trump rivals has a different meaning for some nowadays. =D
They also make their own ARM chips; not as heavily customized as Apple’s, but including their own IP, and with performance competitive with Qualcomm’s best, so not that far from Apple’s (reality distorsion field notwithstanding). Unlike Apple, and like Qualcomm, they not only make the SOCs (the computer bit in the telephone) but the modems too (the part that puts and gets the bits from the air).
And they sell many, many million units around the world. More than Apple, already.
Don’t forget they include a free backdoor with each and every device purchase!
https://thehackernews.com/2016/11/hacking-android-smartphone.html
So does everybody else.
I’m far more concerned about PRISM spying on me than somebody in China spying on me.
Back in the 1990s and 2000s cpu speed jumps were important and noticeable. Nowadays it seems like cpus are so beyond overkill for mundane tasks and generally only power users really notice the bump in performance. So while it gives apple a bonus here, I think many end users would be far more happy with week long battery life instead of having to haul a big battery pack around all the time.
You say that, but I’ve just upgraded my computer from Core 2 level performance to latest core i5 level performance and noticed a BIG step up in performance.
It’s true that each further gain nets less, i.e. diminishing returns. But an upgrade can still be significant, even for just regular stuff like booting your computer, opening and using a web browser, etc. Don’t think simple tasks aren’t speedier because they are simple. There is less delay, not much, but it’s still really noticable.
I think a lot of the speed jumps from a 10 year old CPU to a modern one have to do with cache size and core counts. In my experience, single threaded processes tend to benefit most from higher clock speeds. I’ve had a 5 year old 3.5GHz dual core machine outperform a new quad core 2.4GHz machine at single threaded tasks, but the newer machine stomps the older on multi threaded tasks as expected.
I’ve also seen a measurable (but not quite noticeable) difference between similarly clocked machines with different cache sizes.
Given how most big commercial applications these days are still single threaded, we really won’t see a huge jump in performance until that changes.