Is Apple going to ARM to simply be different?

[leman]

I do not know anything about chip manufacturing, so please apologize if this question is silly... How important are the clocks? I got the impression that the thermals (or performance pro watt) is where AS excels, given that iPad Pros have no active cooling at all.

Higher clocks = better performance. At the same time, higher clocks means higher power usage. The advantage of Apple CPUs is that they can do more work per clock, and therefore can provide good performance at lower clocks and lower power draw. Right now, a mobile Apple CPU running at 2.7 ghz offers similar performance to a 4.5+ ghz Intel CPU (in general-purpose workloads). But Apple needs to do better than that to justify this transition. Getting higher clocks is an obvious way to do it, especially since they can afford slightly higher power consumption on desktop.

Higher clocks, or more cores, or both? As long as you can get enough airflow ?

More cores at similar clocks.

I always wondered that the Intel clocks actually seemed to shrink. My 13" early 2015 MBP has 2,9 GHz, current 13" MBPs have 1,4 (8th gen) or 2,0 (10th gen) without turbo. One of the reasons my machine is still competitive, to a certain degree? BTW, nothing compares to my iPad Pro (1st gen) concerning snappiness.

It’s because the advertised clocks are not much more than marketing. What is important is the actual running clock. Intel has been tweaking the number of cores and the clock range to get more out of their aging architecture. Lowering the base clock while increasing the boost clock is a good way to make a CPU that consumes less energy in sustained multi-core workloads but can still provide bursts of performance for snappiness. But it is not making it easier for the consumers. Which is why I like how Apple is not abusing the turbo clock ranges and instead leverages great energy efficiency and big little architecture to combine power savings and good performance.

 

[leman]

Trust me, I agree. But I also think Apple may do what they did with their switch to Intel. Apple moved from PowerPC 32 to Intel 32 with the Core Duo. It was one year later when Apple went to Intel 64. Since then, there has been nothing but complaining about lack of 32 bit support.

I hope Apple doesn’t do this with a year long half step, like they did with Intel. If they do do that, it would probably be in the GPU.

How would you imagine this situation repeating in the current context? Also, Apple supported 32bit just fine, they only dropped it last autumn. For a while, OS X kernel was 32-bit, with support for 64-bit applications.

 

[vigilant]

How would you imagine this situation repeating in the current context? Also, Apple supported 32bit just fine, they only dropped it last autumn. For a while, OS X kernel was 32-bit, with support for 64-bit applications.

My guess is only a guess. I have no information.

My bet is that Apple has a 5 year roadmap for everything and ML and GPU are the least mature part of it. FPGA is the absolute least part of it.

I think we will see these be the laggards that end up becoming less compatible longer term.

I do think Apple has HUGE plans for the GPU.

If I were Apple, and I was gambling, I’d say that we in the A6/A7 of Apple in the form of GPU’s. My generalized bet is that they’ll fill in the void of things like AVX 512 on the GPU level, but the best way to access that will be through Metal. Apple will need to address any gaps between Intel and AS in the first 2 years.

I fully expect Apple to tell developers to trust the frameworks, and the best path for enhancements and compatibility will be from using those. On the backend, Apple will map out the frameworks to take the “best path forward” which will be a combination of CPU and GPU for the task.

I think first gen AS will suffer for not having as much support for newer hardware level features as a result.

Thats what I think will happen.

 

[JMacHack]

At the end of the day, third-party support relies on two things: marketshare, and how easy Apple makes it for third parties to get involved.

I disagree with this, profit is what drives support rather than sheer marketshare. In terms of phones, Apple is only a majority in the U.S., and can have far lower marketshare in many other places. Yet we see better app support in the App Store than the Play Store because the App Store is far more profitable. The Mac, while not completely analogous to this, sees a similar situation. We see apps on the Mac because Mac users are willing to pay for it, despite being only about 10% of the PC market at a given time.

I don't really know that Apple is going to be blowing Intel and AMD away in terms of performance. Likewise, I don't expect them to exceed AMD or Nvidia on the graphical and specialized computing fronts. Apple's dominance in portable devices won't necessarily translate over to desktop devices (although their laptop lines will probably benefit). I do think that they will be able to start changing the way that a traditional computer operates and is thought of, though. To me, that is the more exciting (and again, frightening) prospect.

I'm not ready to say that they'll blow both X86 giants out of the water either, but I think we'll see them be very competitive. Graphics is up in the air, I agree. But I have a hunch that Apple's got some plan to offer similar or superior performance to what they have now. There'd be no point in changing architectures if they were going to regress in performance. (as a side note, I'm in the "They're gonna keep AMD dGPUs" camp)

I don't think there's reason to be pessimistic about this change, and I'm usually a pessimist myself.

 

[Joelist]

Actually OSX (MacOS) is currently at 17% of the PC Market overall and 27% in the US per StatCounter. Make of that what you will.

 

[leman]

My generalized bet is that they’ll fill in the void of things like AVX 512 on the GPU level, but the best way to access that will be through Metal. Apple will need to address any gaps between Intel and AS in the first 2 years.

They might be closer to it than you think. Apple CPUs can already execute 3x 128bit vector instructions per clock. Newest Intel CPUs can do either 1x 512bit (AVX-512), 2x 256bit (AVX2) or 2x 128bit (SSE). All Apple needs to “catch up” is add another 128bit unit - then they will have the same max throughout and better average case performance. I am curious whether Apple will implement SVE, which will further simplify writing vectorized code. Apple also already supports additional vector extensions (AMX) which are only accessible through Apple frameworks - Intel is planning to bring comparable instructions in
2021. As to GPUs, you can already comfortably use them for parallel processing. It’s just not always the best fit.

Actually OSX (MacOS) is currently at 17% of the PC Market overall and 27% in the US per StatCounter. Make of that what you will.

I’m wondering what is macOS market share among “premium” computers. Probably closer to 50%...

 

[dmccloud]

You need to add together the R&D costs, the costs of the fab, the license cost and the cost of Apple having their own internal hardware design team. Apple Silicon is likely the same or more expensive than Intel for them. This resembles those arguing that Apple created the A Series because it was cheaper while when you add it all up they pay more than if they just put Snapdragons in their products - of course they got a lot of power and fully realized features for their iPhones and iPads and Apple Watches in the bargain.

You are creating a false analogy here. Most of that R&D/production/design budget has already been allocated to A-series development. What this change does is most like result in at worst a scenario where the money and resources that have gone towards working with Intel parts is reallocated into the existing A-series program, resulting in a zero-sum game for Apple. Most likely, the overall costs to Apple actually will be lower, since the costs associated with adding capacity to existing facilities and programs would likely be less than farming out to Intel (and paying their premiums). You mention Apple ditching Snapdragon for A-series, but that's a completely different scenario, as Apple had neither a SoC/processor design team or an agreement with TSMC or another company to manufacture their processors at that time.
Switching from Intel to Apple SoCs does not carry the same type of cost increases that switching from Snapdragon to the A-series had. I'd wager that the total cost to Apple for each processor they manufacture is significantly less than what they pay Intel per CPU when you factor in the fact that Apple will no longer have to design logic boards and hardware for Intel, but can fold that design work into the existing iPad/iOS design processes. If anything, the larger sizes and improved thermal constraints of Apple Silicon Macs will lower those design and development costs compared to Intel-based machines.

 

[dmccloud]

I do not know anything about chip manufacturing, so please apologize if this question is silly... How important are the clocks? I got the impression that the thermals (or performance pro watt) is where AS excels, given that iPad Pros have no active cooling at all.



Higher clocks, or more cores, or both? As long as you can get enough airflow ?

I always wondered that the Intel clocks actually seemed to shrink. My 13" early 2015 MBP has 2,9 GHz, current 13" MBPs have 1,4 (8th gen) or 2,0 (10th gen) without turbo. One of the reasons my machine is still competitive, to a certain degree? BTW, nothing compares to my iPad Pro (1st gen) concerning snappiness.

Part of the reason Apple switched from the 1.4GHz/2.4GHz 8th gen variants in the 2019 MBP 13" to 1.4/2.0 10th gen in the 2020 models is that the 10th gen parts generate more heat, so the only way to accomodate the newer processor without significantly altering the internals and cooling system was to go with a lower clocked model. Also, clock speed by itself is a poor indicator of performance. For example, the 8th generation core processors had on average 20% better performance than their 7th generation predecessors. In short, an 8th gen i5 running at 2.0GHz had performance comparable to a 2.4GHz 7th gen i5. This is why the performance per watt metric has become increasingly important.

 

[Boil]

This is why the performance per watt metric has become increasingly important.

Over on the SFF.Network forums we have a metric we like to call the PPL, Performance Per Liter...

Which is why I would LOVE to see a return of the Cube...!!!

 

[jerwin]

The next step up from ARM ISA based SoCs is to run Swift natively. They are far from it, but more than one Apple beat type reporter has stated that is the next transition. And when I say Swift natively, I mean Swift will be the instruciton set, nothing below it.

Just like the Intel 432

 

[jerwin]

Hey don’t diss the CASIO PB-1000 with 16 LCD touchscreen keys... iPad Pro eat your heart out.

I remember those days of "who need a mouse and keyboard when you can just use a touchscreen. " It really doen't work when the system you are trying to replace is an S/370 based library cataloguing system

 

[wardie]

Over on the SFF.Network forums we have a metric we like to call the PPL, Performance Per Liter...

Which is why I would LOVE to see a return of the Cube...!!!

Ah nice link I see it’s performance per litre of chassis space, not what I thought... per litre of coolant pumped! (Sorry, UK spell checker there)

 

[Ledgem]

I do not know anything about chip manufacturing, so please apologize if this question is silly... How important are the clocks? I got the impression that the thermals (or performance pro watt) is where AS excels, given that iPad Pros have no active cooling at all.
...
I always wondered that the Intel clocks actually seemed to shrink. My 13" early 2015 MBP has 2,9 GHz, current 13" MBPs have 1,4 (8th gen) or 2,0 (10th gen) without turbo. One of the reasons my machine is still competitive, to a certain degree? BTW, nothing compares to my iPad Pro (1st gen) concerning snappiness.

It used to be that the clock rate meant everything. I still remember the days when advertisements said things like "be the first on your block to own a 1 GHz CPU!" Ah, how far we've come.

It's become more complicated these days, even going beyond the simple comparison of clock speed vs cores. As a general rule, higher clock rates are better, as it allows the computer to perform more tasks in a given amount of time. But there's more to processor design than that. If the computing task that is being done can be broken up and done in parallel, having more cores - even if they're clocked more slowly - can match and exceed a system that is running at a higher clock rate. Because higher clock rates generate more heat, and system stability also relies on running within a certain temperature range, systems with a lower clock rate may come out ahead there as well. This is partly where Intel's "turbo boost" technology is also rather impressive; you can ramp up the clock rate when you need it, and keep the system running a bit cooler when you don't.

But it also depends on what you're trying to do, and what specialized functions your hardware may have. As an example, consider video decoding. Large videos using newer codecs can be choppy and unwatchable on older computers, and you'd see the CPU usage maxing out to try and accommodate it. Try the video on a newer computer, and it plays smoothly - and with barely any increased usage from the CPU. What's going on? In those cases it may be that the CPU either has a specialized instruction set, or that there's something else helping (the graphics card, or Apple's T2 chip) to offload the task from the CPU.

The model of one single CPU for everything seems to be fading in favor of having more specialized processors and/or processor cores. It seems a fair bit of suspicion is that this is the direction Apple will go with their own chips. It seems to be more efficient, meaning less energy consumed and thus less heat generated, too.

 

[MisterMe]

The next step up from ARM ISA based SoCs is to run Swift natively. They are far from it, but more than one Apple beat type reporter has stated that is the next transition. And when I say Swift natively, I mean Swift will be the instruciton set, nothing below it.

This would not be a step forward. To the contrary, it would be a huge step backward. jerwin referenced the "intel 432." He is referring to Intel's iAPX 432, a processor intended to be the first to be coded in a high level language. The iAPX 432 had been in development since 1975. However, it was supposed to be coded in Ada, the Department of Defense's mandated language. For a hot minute, the iAPX 432 was a really big deal. However, its lost its big deal status log before Ada lost its Federal mandate.

 

[Kostask]

that was a 2 package CPU, from what I can recall. 2 really big packages, and needed a couple of other large ICs to be usable. Massive failure, and I think it was originally intended for military applications, or to get Intel into the mini/mainframe world (Vax/IBM370 level machines)

 

[/V\acpower]

I think you've got it backwards. Just because many don't think of it as a walled garden doesn't mean that it isn't one, or that it doesn't matter. The walled garden approach is something that continues to hobble the iPad because no matter how powerful the hardware gets, it is inherently limited by what the software allows users to do. If we limited ourselves to simply making things to fit the masses, we'd never have anything good, because the masses are largely unimaginative or uninformed.

I don't think the iPad is a fair comparison. The real problem of the iPad is not that every app has to come from the App Store, but wanting to have a powerful computer who can work only with Touch.

If you had a Mac with the same "keyboard + trackpad/mouse", windows, menu bar, etc., even with being only being able to run Apps from the App Store, it would be still be miles ahead of the iPad in term of productivity for a whole lot of users.

Basically, it's not the App Store that make the iPad a bad productivity device.

However, that being said, I don't think that Apple should and will close the Mac the same way they do with iOS. The Mac is a development platform, being able to run whatever code you want is essential to the Mac. It's the platform for peoples who accept to use a device that you have to "manage" more.

I also don't believe the reason iOS is "closed" is simply because of greed. Fundamentally I believe they still think it's the kind of thing that make a better user experience for the kind of user they are aiming for with the iPhone and iPad. And I think that it would make way less sense for them to do the same on the Mac because they are not aiming for the same kind of users.

 

[Kostask]

This would not be a step forward. To the contrary, it would be a huge step backward. jerwin referenced the "intel 432." He is referring to Intel's iAPX 432, a processor intended to be the first to be coded in a high level language. The iAPX 432 had been in development since 1975. However, it was supposed to be coded in Ada, the Department of Defense's mandated language. For a hot minute, the iAPX 432 was a really big deal. However, its lost its big deal status log before Ada lost its Federal mandate.

I also remember way back when, Western Digital made a CPU/Chipset that ran Pascal intermediate code (p code) natively instead of assembly code. It was in the late 1970s/early 1980s, and was very expensive, and a very big deal at the time. It died when Pascal started to fade away. It was not that fast, either.

I'm pretty sure that silicon technology has moved along quite a bit since the late 1970s/early 1980s. Its also quite a bit faster. I don't see running Swift natively as being as big a reach as most here make it out to be. Instead of thousands or tens of thousands of transistors, it is common to see billions of transistors, sometimes over 10 billion, and instead of running at 1-8 MHz, CPUs are running well into the GHz, 2-3 GHz in the case of the Apple SoCs.