Apple Plans to Ditch Intel and Use Custom Mac Chips Starting in 2020

[cmaier]

Uh huh.

Apple did not make chips in the 90's; they were part of a consortium that included IBM.

In the end, even the combined resources of the consortium could not keep pace. Apple gave up so hard that they had to screw over their users and developers with an architecture switch.

Now Apple will go it alone and somehow beat Intel in Intel's market. In a way that AMD, IBM, Qualcomm, Samsung, etc. never could.

Do you really think that's true, or do you think that Apple will do what Apple is good at? Produce a demonstrably mediocre product but use their insane marketing engine to convince consumers that mediocre is the best.

I will not bet money that Apple is going to somehow become the best processor maker in the world just by wishing that they are.

I bet they will convince themselves and their users that those so-called "high-power" chips just aren't needed.

Apple didn't design CPUs in the 90's. Now they do. They've hired some of the best designers in the business, many of whom I know personally and were my former colleagues , who regularly have beaten intel at their own game in the past. They also now have the volumes to compete with Intel and thanks to TSMC and samsung there is little Intel fab advantage anymore. It was the fab advantage that killed powerpc, not bad designs.

 

[dirt_farmer]

Most people here aren't old enough to remember the Megahertz Myth.

This was trotted out by Apple at every presentation to demonstrate that clock speed is not a good metric of processing power, and instead it's important to consider benchmarks and real-world use.

They did this because their clock speed was lower but RISC architecture made up for it by being more efficient.

Apple reminded everyone of this right up until their Power series processors started being slower by *every* measure, not just clock speed.

Then Apple went silent and focused on how Macs are better for other, intangible reasons -- despite having weak and anemic silicon.

That's my prediction for Apple in 24 months. "Don't worry about the fact that our new chips suck; think about how much you like using Facetime".

Yawn.

 

[cmaier]

That is exactly what I have been saying since the thread started.
A11 is not a competitor to an i9, no matter what Geekbench says.

From Geekbench
Nov 16, 2017 iMac19,1Intel Core i9-7900X 3312 MHz (10 cores) Mac OS X 64-bit 5644 42660
Oct 26, 2017 MacBook Pro (15-inch Mid 2017)Intel Core i7-7920HQ 3100 MHz (4 cores) Mac OS X 64-bit 4969 16999
Apr 04, 2018 iPhone XApple A11 Bionic 2390 MHz (6 cores) iOS 64-bit 4262 10529

Now Geekbench gives a number for single and multicore workloads.
They claim real world workloads but I haven't seen the code.
But normalizing the multicore performance and just scaling speed; that gives
(1.29 x 10529)/6 = 2263 - A11
(1.06 x 16999)/4 = 4504 - i7
42660/10 = 4266 - i9

Clearly that's not how we evaluate performance but even normalizing for speed; the multicore performance of the A11 needs double the performance per core to be in the class of an i7 or i9.
For a single core normalizing for clock speed gets you in the ballpark, but that makes a lot of assumptions and none of them include the I/O for a balanced laptop/desktop system that the A11 does not have.

Once again, I didn't say Apple can't do it.
The ROI needs to make sense to invest the billion dollars in people, tools, silicon characterization, packaging research, etc. that it's going to take to even start to get a family of processors.

Ask Sun, HP, HAL and others how much money it takes to go head to head with Intel.

Remember that they get a large boost in performance without changing anything simply by putting the cpu in a laptop/desktop thermal solution. Geekbench runs cause extensive throttling of mobile cpus in mobile devices because of thermal considerations.

 

[dirt_farmer]

Apple didn't design CPUs in the 90's. Now they do. They've hired some of the best designers in the business, many of whom I know personally and were my former colleagues , who regularly have beaten intel at their own game in the past. They also now have the volumes to compete with Intel and thanks to TSMC and samsung there is little Intel fab advantage anymore. It was the fab advantage that killed powerpc, not bad designs.

I hope you're right.

Having smart designers isn't always enough -- as Apple has spent 5 years painstakingly demonstrating via almost every product line.

They have the best hardware designers on the planet and have switched their design language from "category-defining" to "dated also-ran".

Same with software come to think of it.

https://www.cnbc.com/2018/01/30/apple-is-reportedly-delaying-new-ios-features-until-next-year.html

Smart designers is just one step out of ten thousand on the road to shipping a good product. Never mind the world's best product which is head and shoulders above the combined efforts of an entire industry.

Again, though, I hope you're right.

 

[cram501]

Only the marginal cost of arm chips will be cheaper. They have to pay for the R&D. It may actually cost them more in the end. but the resulting chips will be more powerful than Intel's garbage, burn less power, and will actual advance year-to-year instead of being stagnant for generations. We kicked Intel's ass in the 90's by using a simpler architecture and Apple is going to do the same thing with the same chip designers.

Who kicked Chipzilla ass in the 90's? I really don't see anyone else around except some recent stuff by AMD. Intel deserves to be kicked a little for being so complacent without any real competition in the last decade. Their chips contributed to Apples success. It didn't hinder it. Intel should be congratulated by Apple fans and not derided.

Until Apple proves they can play with the big boys, it's all vaporware and hope. Developing a desktop chip is lot more difficult and takes a lot more time and money than people seem to realize. If it wasn't, we would have a lot of choice already.

 

[ssgbryan]

Most apps require only recompile and slight tinkering. If devs are lazy it's on them, because things change sooner or later. We also have no idea how relevant windows is to majority of mac users.

Really?

You remember the transition from PPC to Intel very differently than I do.

It took years for software houses to move my mission critical software.

 

[techwhiz]

Remember that they get a large boost in performance without changing anything simply by putting the cpu in a laptop/desktop thermal solution. Geekbench runs cause extensive throttling of mobile cpus in mobile devices because of thermal considerations.

But you run into the same issue with desktop CPUs as workloads increase if you have thermal issues.
I don't think you will double the performance by adding a slug, heatsink, etc.

But looking at Qualcomm and Cavium is an example of ARMv8 competing against Intel.
But once again, they aren't scaled up mobile chips but server class chips from the ground up.
That's the point I've been trying to make.

Can Apple do it? I have no doubt that they can.

But I'll go back to my previous points.
1. I have yet to see the hiring of the types of people it will take to get it done here in the valley or in Austin.
You only have two places where you are going to hire the talent in numbers great enough to get this done.
There is a finite pool of CPU and ASIC/SoC talent at the experience levels to get this done. You have to hire them from somewhere.

2. I haven't seen any announcements from those that supply fab capacity about long term agreements.
TSMC typically won't do those agreements and makes everyone get in line. Apple tried an agreement with TSMC for guaranteed capacity and TSMC said no. Apple isn't building a fab. I haven't seen any up for sale. Not at 10nm anyway.
AMD uses Global. But they got their agreement because they sold fab assets to Global. IBM sold their commercial foundry business to Global. Maybe Apple goes to Global Foundries?

For me it's about the logistics and timeline.
I know what it takes to make chips and replacing Intel in two years as touted by Bloomberg doesn't seem realistic.
These are the same folks that said the same thing two years ago or so. They are also the same group that said Apple would be using their own modems by now and would have switched from Qualcomm and Intel.

Bloomberg analysts just don't know jack about the silicon industry.

 

[ssgbryan]

Nice soundbite, but care to come back with a specific point of contention about what this will disrupt for you, or is this just about grumbling because you've become used to a specific feature that's possibly going away? There's plenty of perfectly good windows machines that can be had for little more than the cost of upgrading a MBP to 1TB of storage (pretty much the minimum if you want a functionally large windows partition with space enough left over on the mac side).


Does your programming job really have you on the go so much that having a dedicated windows computer isn't easier than rebooting into Windows? You could get two cheap Windows computers for the cost of upgrading a MBP to 1TB of storage. Have one at home and one at your office, if you get a laptop then take it with you when you need to go somewhere.

2TB wouldn’t even hold my iTunes library, much less what I use my computer for (3D art).

All of my software is designed to take as many cores & ram as I can throw at it. ARM isn’t an option for those that create content (as opposed to those that just consume content).

As a Mac user, I am already planning an exit strategy.

I need a computer, not an iToy.

 

[turbineseaplane]

Well..

Maybe this will make it easier for me to convince myself to go buy the ThinkPad I've been wanting ever since they changed to the 2016/2017 MBP designs..

 

[techwhiz]

2TB wouldn’t even hold my iTunes library, much less what I use my computer for (3D art).

All of my software is designed to take as many cores & ram as I can throw at it. ARM isn’t an option for those that create content (as opposed to those that just consume content).

As a Mac user, I am already planning an exit strategy.

I need a computer, not an iToy.

Which is why I still have my MacPro 2008 with eight Xeons and stuffed to the gills with memory and about 12TB f network attached RAID storage.
They can call me when they have a new pro machine with some slots.
The can idea is as bad as the Mac Cube -> https://en.wikipedia.org/wiki/Power_Mac_G4_Cube

Apple needs to stop playing and give me a pro machine with some internal expansion.

 

[dirt_farmer]

2TB wouldn’t even hold my iTunes library, much less what I use my computer for (3D art).

All of my software is designed to take as many cores & ram as I can throw at it. ARM isn’t an option for those that create content (as opposed to those that just consume content).

As a Mac user, I am already planning an exit strategy.

I need a computer, not an iToy.

THIS.

The departments at my company that process data (Accounting/Finance and Engineering) need components for which their tools have been a) written+compiled and b) optimized.

That means Quadro cards ONLY and x64 ONLY.

The vendors would have us upload all data to the cloud and rent processor cycles to do the crunching...this is a nonstarter.

Computing as a service is being set back by the Facebooks and Cambridge Analyticas of the world.

So we buy workstations that contain workstation processors and workstation graphics and run them h-a-r-d.

Not sure how Apple will maintain the "Pro" moniker with a straight face if they switch to chips that professional software can't use or can't use well.

Oh, wait, they already have an OS like that. Reality Distortion Field for the win!

 

[turbineseaplane]

Not sure how Apple will maintain the "Pro" moniker with a straight face if they switch to chips that professional software can't use or can't use well.

I'm not really sure how they do that now to be honest.. (thinking of laptops in particular)
lol

 

[ssgbryan]

Which is why I still have my MacPro 2008 with eight Xeons and stuffed to the gills with memory and about 12TB f network attached RAID storage.
They can call me when they have a new pro machine with some slots.
The can idea is as bad as the Mac Cube -> https://en.wikipedia.org/wiki/Power_Mac_G4_Cube

Apple needs to stop playing and give me a pro machine with some internal expansion.

I hear you. I am on a flashed 4,1. Next upgrade is CPUs (2.26 quads to 3.06 hexes), replacing 4 2TB HDs with 4 6TBs, and trying to score a modern GPU at a reasonable price.

 

[DNichter]

App choice better on iPad? You can run decades worth of software from Mac/Windows/Linux on the Mac... the iPad is pretty well limited to what's currently available on the App store (filtered by Apple) and supported for your (permanent) version of iOS.

Many of the Apps people use on iOS are just tailored versions of their desktop browser websites to make it easier to use with touch + small screens.

Yes, much better variety of quality apps for the iPad. I am able to find an app for almost any task I need to handle.

 

[Hps1]

What tasks are those exactly? It appears from your posts that you don't need a desktop computer at all.

Edit: wasn't meant to be offensive. I just believe the iOS realm meant for one type of consuming and creation and macOS a complete other.

Unless there are deep learning or realtime raytracing apps on the iPad now...

 

[DNichter]

No, some people's workflows depend on things that iOS can not do. Some tasks will never be achievable on iOS due to the limitations which make it work for other people. Please recognise that people have different requirements and workflows, that are not really that easy to adjust. It is not about laziness.

And yes, bringing iOS to the Mac would ruin a Mac. There is no point turning the Mac lineup into giant iPads... because Apple already sell iPads.

Again, if the iPad works for you, then I'm glad. How about you recognise that the Mac also works for others.
[doublepost=1522859916][/doublepost]

And your idea of a better OS seems to be one that is locked down like iOS, so essentially would be iOS, thus ruining the advantages that the Mac has over iOS devices.

I am aware, the Mac isn't going anywhere. It's just being improved upon. Apparently Apple agrees.

 

[dirt_farmer]

I'm not really sure how they do that now to be honest.. (thinking of laptops in particular)
lol

You mean since the most powerful "Pro" laptop they make has a year-old Ultrabook processor?

Since a portable workstation from any tier-1 vendor can return benchmarks that are literally ten fold what a Macbook "Pro" does?

If I was still a Mac user, I would be MAD.

 

[deconstruct60]

Apple didn't design CPUs in the 90's. Now they do. They've hired some of the best designers in the business, many of whom I know personally and were my former colleagues , who regularly have beaten intel at their own game in the past.

Apple has been competitive in the CPU game because they have been laser phone focused on one area. Not trying to be everything for everybody.

The beat Qualcomm to the punch on 64 bit transition because they were not looking at 64 as a server thing and only has better/cleaner opcodes thing.

They also now have the volumes to compete with Intel and thanks to TSMC and samsung there is little Intel fab advantage anymore. It was the fab advantage that killed powerpc, not bad designs.

Eh? TSMC and Samsung have capacity (and process implementations that as slowing across the board ) not volumes. The Apple iOS ( iPhones in particular) have volumes ( several 10's of millions per year creeping close to 100M threshold). The Mac doesn't (just a bit over 10M threshold). That is even more disparate when you separate out the different sub-segments of laptop and desktop. Every single iPhone in a specific year gets the same processor. Apple's Mac line up has at least 3 separate dies and about triple that in processor product SKUs.

Apple could cherry pick of some low end laptop to put a single ARM implementation into but that would most likely be in the single digit millions range. Pretty close to being about two orders of magnitude different than the iPhone volume.

At best what Apple might do is take a ARM implementation optimized for a future iPad Pro and toss that into a low end (relatively to average Mac prices) laptop. However, replacing the whole line up.... there is very little there to talk about when it comes to "volume". The double digit million iPad would primarily support the chip and the Mac would slurp up secondary usages closer to the way the AppleTV and HomePod do now of the iPhone line up.


IBM and Sun/Oracle make Power and Sparc competitor processors are relatively sub 1M run rates, but Intel is eating their lunch with more cost effective options. The very top end of the Mac line up is barely going to be in the 1M range. If Apple chose to dump that, they could but to fill that last upper 20% the cost benefit analysis gets highly dubious. It works for Intel (and AMD) because they have several system vendor customers to spread the R&D costs over. If Apple does it all by themselves, they have no critical volume in the top 20% of the line up.

 

[Sydde]

Threads in hardware or hyper threading. I was not talking OS.

Hyperthreading is just an MP strategy. Its value is debatable. The ARM group suggested that it can significantly increase cache thrashing, which is a bad thing that can reduce performance considerably (memory access is comparatively slow, so forcing the L1 into fetch/dump/refetch/redump/refetch cycle is very costly). A73 has good OoOE, which can be enough to make up the difference in full utilization of processor resources, and apparently has a very good dynamic branch predictor that keeps the fairly short pipeline mostly free of bubbles.

In other words, HT is not a crucial performance need. And as more heavy work gets offloaded to things like the (GP-)GPU, the actual importance of CPU core performance becomes less critical.

 

[deconstruct60]

Remember that they get a large boost in performance without changing anything simply by putting the cpu in a laptop/desktop thermal solution. Geekbench runs cause extensive throttling of mobile cpus in mobile devices because of thermal considerations.

Yes but that larger thermal solution is probably going to leak back into implementation too. More extensive thermal monitoring and fine grained sleep control with wider ranges between asleep and "full speed".

These geekbench runs aren't that long if I recall correctly. 1-3 minutes. If throttling in 1 minute how can the implementation be optimize to run for hour or so? Apple's modern A-series has asymmetric cores. (e.g. 2 big + 4 small) because the workloads are generally spotty and most of the time it is a "race to sleep".

The A-series does better with Geekbench because it only last a minute or two. Apple has done a lot to optimize the implementation for the specific workloads that phones present. This new implementation they green lighted seems more likely to be a slightly bigger fork for future iPad Pro than a Mac. The iPad Pro has concurrent Apple now ( is side by side windows) , an app dock so quicker transitions between apps , etc. It is taking on more workload that would have been exclusively entry level Mac. So it would make sense to separate that from the phone optimized core over the next 2-4 years.

However, that won't get you something that covers the whole Mac line up.

 

[cmaier]

But I'll go back to my previous points.
1. I have yet to see the hiring of the types of people it will take to get it done here in the valley or in Austin.
You only have two places where you are going to hire the talent in numbers great enough to get this done.
There is a finite pool of CPU and ASIC/SoC talent at the experience levels to get this done. You have to hire them from somewhere.

When we designed Opteron with a MUCH smaller team than Apple already has (there were maybe 20 key folks dedicated to it). Apple has already hired a large chunk of the folks who worked on those chips on AMD, folks who worked on DEC's Alphas, and folks who worked on the Exponential PowerPC. Plus many other people. They have more than enough talented people to get it done.

 

[inkswamp]

The thing is it's, as you say, a nice "convenience" now. I'm sure Apple doesn't want it's customers buying Microsoft Windows licenses and lets face it the amount of PC gamers buying Apple hardware to run games is absurdly small and Apple won't loose any sleep turning it's back on them.

I didn't say anything about PC gamers buying Macs to run Windows. It's more of a selling point for Apple, that customers know they have the option to run Windows as well. I suspect Apple customers who rely on this "convenience" is not as absurdly small as you make it out to be.

Besides, I'm trying to figure out what the benefit is for doing this. What does Apple gain by bringing Mac OS and iOS closer? Running touch apps on a Mac? Running Mac apps on touch devices? Gross. We've seen this done by Microsoft and it's a freaking mess as well as the kind of UI nightmare Apple avoids like the plague. Until someone can explain to me a significant upside to doing this, I will continue to view losing compatibility with the Intel world as a huge loss for Apple and its customers.

 

[cmaier]

Eh? TSMC and Samsung have capacity (and process implementations that as slowing across the board ) not volumes. The Apple iOS ( iPhones in particular) have volumes ( several 10's of millions per year creeping close to 100M threshold). The Mac doesn't (just a bit over 10M threshold). That is even more disparate when you separate out the different sub-segments of laptop and desktop. Every single iPhone in a specific year gets the same processor. Apple's Mac line up has at least 3 separate dies and about triple that in processor product SKUs.

I don't understand any of this. Grammar issue? Apple buys millions upon millions of die. In the 90's, they needed, what, 5 million powerpc's a quarter? The whole powerpc demand was maybe a few times that? That's not enough to support billion dollar fabs and pay for the researchers needed to push the technology. Now they do enough volume to keep up.

 

[deconstruct60]

But I'll go back to my previous points.
1. I have yet to see the hiring of the types of people it will take to get it done here in the valley or in Austin.

there are places with talent outside of Austin and the valley. Cheaper too. However, one thing that seemed to sorely missing from the Bloomberg article is that other devices that Apple is trying to push ARM into also how much dilution they will get from their team chasing those too.

One of the critical things they'll have to do is get volume. Being able to use one CPU design you whole Apple Watch line up scales much better than a Mac line that needs 2-3 different dies designs at substantially lower volumes.

There is a critical issue how Apple keeps ahead in the area where they are already ahead. How do they keep their phone targeted implementation ahead of the game if they start to stick their fingers in the 4-5 different pots.

If Intel dropped some of the SKUs in their product line they'd probably be better. That's part of the issue is that they focus is divided. Apple hasn't had divided and that has highly contributed to them being in front.

At some point hit the "Mythical Man Month".... we can get the software out faster and in less time if just add more bodies. It is similar for all projects that involve high internal complexity and coupling. More people isn't going to necessarily make it all better.

I know what it takes to make chips and replacing Intel in two years as touted by Bloomberg doesn't seem realistic.
These are the same folks that said the same thing two years ago or so. They are also the same group that said Apple would be using their own modems by now and would have switched from Qualcomm and Intel.

Bloomberg analysts just don't know jack about the silicon industry.

I wouldn't go so far as to say they don't know jack, but this is an old Mac Rumors writer. These articles of late about Apple read more like clickbait Macrumors front page stories that careful technical analysis. Bloomberg has ads too so the clickbait aspects and the "sky is falling what Intel stock drop" on Bloomberg TV only helps pay the bills.

I don't think there is an assertion that they would do a one year flip of whole line up buried in the article. I suspect Apple is looking at some of the "Windows on ARM" activity and the projections for what they need to make iPad competitive ( from now they can keep thinking that Chromebooks aren't eating their lunch in edu ).
[doublepost=1522870885][/doublepost]

I don't understand any of this. Grammar issue? Apple buys millions upon millions of die.

Right, to use in one device product. iPhjone 8 , iPhone 8Plus , iPhone X all same SoC. They don't have different ones.
iPad ... same SoC across that years model ( or some already reached cost recovery one like the "new" iPad 2018 with two year old SoC. ).
Apple Watch. one SoC all of the this product.

Mac .... not.

In the 90's, they needed, what, 5 million powerpc's a quarter?

The PowerPC lost. ( yes there is some elements left in the modern Power ). 5 million won't cut it. That is a looser volume up against what Intel+AMD are doing in terms of revenue generating to keep supporting pushes to more and more expensive process tech.

The whole powerpc demand was maybe a few times that? That's not enough to support billion dollar fabs and pay for the researchers needed to push the technology. Now they do enough volume to keep up.

Fabs cost 10's of Billion these days.

 

[Altis]

Yes, much better variety of quality apps for the iPad. I am able to find an app for almost any task I need to handle.

The difference is you need to find an app for almost every task. It's a song and a dance just to do what's so easily done with a desktop OS.

>> Oh, you want to do this simple little task? Well first you need to have an Apple account and download/buy these 12 apps!