Why didn't Apple have Intel manufacture their chips?

[willfr]

The more I think about this the more it seems like Apple made a bad call by not having Intel manufacture their chips. They had an existing and good relationship with Intel since macs switched to intel. All iDevices would have been significantly better for users (and crucially, significantly better than competing Android phones). Imagine if the A10 was built on Intel new 14nm+ process? Apple has been leaving a lot on the table for years now by not having Intel manufacture their chips. It's a shame really

 

[Nothingness]

First you're making the assumption Intel process is technically superior. There's no proof of that yet. I guess Intel is better but not that much better that it'd make a significant difference.

Second you're making the assumption Intel was willing to fab chips for Apple back then. In fact Intel said no to Jobs when he asked them.

Third you're making the assumption that Intel was willing to fab mass market non-x86 chips. This only changed recently.

And fourth porting to a new process is not easy, especially when the foundry isn't very experienced at supporting many customers, as is the case with Intel Custom Foundry.

 

[itsmydamnation]

fifth There is almost no 3rd party IP ready and available on intel

 

[ShintaiDK]

Because Intel as such was a competing company until they dropped the mobile Atom dreams. As they dropped that you saw a big change right away. LG coming on, ARM cooperation etc.

Apple will be a future Intel custom foundry customer.

 

[Qwertilot]

Perhaps. TSMC have been a third party foundry for much longer than Intel and they do seem to be awfully good at it. Apple do also obviously really like multi sourcing though.

 

[Arachnotronic]

fifth There is almost no 3rd party IP ready and available on intel

This is a really big one and I think one that a lot of people don't appreciate. Modern SoCs rely on a lot of external IP sourced from the likes of Synopsys, ARM, etc. The SoC vendors' job is to build key differentiating IP + good system fabric + software more than anything else.

 

[Arachnotronic]

The more I think about this the more it seems like Apple made a bad call by not having Intel manufacture their chips. They had an existing and good relationship with Intel since macs switched to intel. All iDevices would have been significantly better for users (and crucially, significantly better than competing Android phones). Imagine if the A10 was built on Intel new 14nm+ process? Apple has been leaving a lot on the table for years now by not having Intel manufacture their chips. It's a shame really

14nm+ is nice for high performance MPUs and from the data I've seen published, Intel's transistors tend to really thrive at higher leakage current levels than they do at much lower leakage current levels.

If you are building a server processor, 4GHz+ desktop CPU, etc. there's no better process than 14nm+ in the market today. If you are building a low power SoC, well...when Spreadtrum starts shipping its 14nm SoCs and we get a look at xtor density, performance of the xtors, etc. then we can see if Intel really has much of an advantage for mobile SoC class hardware.

 

[Phynaz]

Give it a couple of years for Apple to have designs developed to Intel's process rules.

 

[Dribble]

Money - Intel expect huge margins for chips they make and Apple expect to keep all the margin for devices they make (i.e Apple are used to their chip provider producing chips practically at cost). Tough to negotiate a deal there.

 

[ShintaiDK]

Apple already pay out the nose for their chips.

 

[Arachnotronic]

Money - Intel expect huge margins for chips they make and Apple expect to keep all the margin for devices they make (i.e Apple are used to their chip provider producing chips practically at cost). Tough to negotiate a deal there.

TSMC's average margins are like 50%+.

 

[FIVR]

Intel has been trying its hardest to get Apple to fab something with them for years, the idea that they were "too good" or "too competitive" and thus threw out billions of dollars in revenue is ridiculous on its face. It isn't surprising given the caste of characters on this forum that such blatant falsehoods are plastered all over the boards, but I guess that's just the way things are here.


Intel's process has some slight advantages to TSMCs, but intel also has huge downsides such as lack of in house IP and zero experience doing anything for anybody but themselves. The advantages intel has are the electrical characteristics of their transistors, but their density actually appears to be inferior to TSMCs 16FF+ and especially Samsung 14FF. They've also been attempting to fab Altera FPGAs since 2014 with little success. Given these facts, it is obvious why Apple has chosen to go elsewhere.

OP, be careful what you wish for. The intel modem they managed to jam into a few unfortunate iPhones is actually inferior to Qualcomm's competing version (it is CAT10 450Mbps while Qualcomm has CAT12 650Mbps capabilities), this has already resulted in poorer performance on the new iPhone 7. I'm pretty sure Apple will dump them next year, but for now Apple has chosen to artificially limit Qualcomm modems to CAT10 in order to have uniformity in their product. So, adding intel parts to the iPhone did nothing but make it perform poorer. Why would you want an intel SoC to slow your iPhone down?

 

[witeken]

Plot twist: why doesn't Apple manufacture its own chips?

First you're making the assumption Intel process is technically superior. There's no proof of that yet. I guess Intel is better but not that much better that it'd make a significant difference.

Comepletely wrong. Intel proved their 14nm process is 40% denser than BOTH Samsung and TSMC a year ago.

Don't confuse you ignorance (not knowing stuff) with the actual facts.

 

[witeken]

Give it a couple of years for Apple to have designs developed to Intel's process rules.

This a thousand times.

Just look at it this way: Apple -- or so the story goes, anyway (I'm just repeating what people like AnandTech say) -- likes to have diversity. They (just like most other companies, for instance Intel for their tools) don't want to be dependent on one company. So Apple can design their stuff for TSMC, Samsung, Intel, they don't really care 'cause they have enough money, and they will just go with the company that can meet their demands for THAT specific product.

 

[Nothingness]

Comepletely wrong. Intel proved their 14nm process is 40% denser than BOTH Samsung and TSMC a year ago.

Don't confuse you ignorance (not knowing stuff) with the actual facts.

Don't confuse your bias with knowledge. There are several sides to performance. Area is one of them. I leave you as an exercise to list the others

Give it a couple of years for Apple to have designs developed to Intel's process rules.

That's indeed quite likely, things have changed a lot. But I doubt they'll ever use Intel as a single provider for anything, be it modem or foundry.

 

[witeken]

Don't confuse your bias with knowledge.

Don't confuse my profile picture with bias. I don't have time these silly I-support-company-X-and-company-Y-is-poor games.'

For sure I can only base my opinions on what I know, but I'm always open for new information.

 

[Arachnotronic]

Plot twist: why doesn't Apple manufacture its own chips?


Comepletely wrong. Intel proved their 14nm process is 40% denser than BOTH Samsung and TSMC a year ago.

Don't confuse you ignorance (not knowing stuff) with the actual facts.

Pretty sure Nothingness is an actual chip designer...

 

[witeken]

Pretty sure Nothingness is an actual chip designer...

Come on. Denying Intel's 14nm process' superiority clearly shows that you're missing something, which is strange given he's been on this forum for so long.

But if he could discuss with actual information, then I'm fine to be proved wrong. (It really boggles my mind that when people make claims they never refer to actual information. In the literature, this is called plagiarism, or else your facts are just made up, because I doubt it's own research.)

 

[carop]

Give it a couple of years for Apple to have designs developed to Intel's process rules.

- Transition from x86 to ARM.

- Pit Intel Custom Foundry against TSMC for manufacturing the chips.

- Lower the price of Macs to reduce the Mac tax and use iOS to make inroads into the Enterprise market which has the potential to be huge.

 

[carop]

double post sorry!

 

[Arachnotronic]

Come on. Denying Intel's 14nm process' superiority clearly shows that you're missing something, which is strange given he's been on this forum for so long.

But if he could discuss with actual information, then I'm fine to be proved wrong. (It really boggles my mind that when people make claims they never refer to actual information. In the literature, this is called plagiarism, or else your facts are just made up, because I doubt it's own research.)

14nm process superiority is a claim made by Intel. In the real world, for whatever reason, very few people are using ICF and the bulk of chip designers are using TSMC 16FFP/FFC and soon TSMC 10nm.

There is more to it than technology obviously, but outside of high performance CPUs where Intel's products/process show a demonstrable lead, there's no evidence to suggest that Intel's 14nm is better than TSMC's 16nm for mobile chips, all things considered.

 

[witeken]

14nm process superiority is a claim made by Intel. In the real world, for whatever reason, very few people are using ICF and the bulk of chip designers are using TSMC 16FFP/FFC and soon TSMC 10nm.

There is more to it than technology obviously, but outside of high performance CPUs where Intel's products/process show a demonstrable lead, there's no evidence to suggest that Intel's 14nm is better than TSMC's 16nm for mobile chips, all things considered.

I am indeed not aware of quantitative comparisons like mv/dec (but Intel's 14nm process is close to the silicon theorectical minimum of 60mv/dec with its 65 value) or other power/frequency things that are apples to apples. But we know that feature size is still important, and those are superior just like other characteristics (gate length, air gaps, fin pitch, novel subfin doping technique, variation).

http://www.intel.com/content/dam/ww...foundry/intel-14nm-iedm-2014-presentation.pdf

 

[itsmydamnation]

This is a really big one and I think one that a lot of people don't appreciate. Modern SoCs rely on a lot of external IP sourced from the likes of Synopsys, ARM, etc. The SoC vendors' job is to build key differentiating IP + good system fabric + software more than anything else.

In Aaron spink's opinion it is almost the only reason that Intel has struggled the crack into foundry market, not price or design rule restriction or anything like that. Its a brutal chicken and egg cycle to break out of.

 

[IntelUser2000]

I am indeed not aware of quantitative comparisons like mv/dec (but Intel's 14nm process is close to the silicon theorectical minimum of 60mv/dec with its 65 value) or other power/frequency things that are apples to apples. But we know that feature size is still important, and those are superior just like other characteristics (gate length, air gaps, fin pitch, novel subfin doping technique, variation).

http://www.intel.com/content/dam/ww...foundry/intel-14nm-iedm-2014-presentation.pdf

I've argued for years only products matter. Fancy presentations, CEO hyping up the product, reputable but otherwise retired CPU designer, all really don't matter. People are biased. They can also be wrong.

Since 22nm Intel was FULL of claims that didn't pan out in reality. 37% faster with 22nm Tri-Gate transistors. Where? Later we find out that Silvermont(Atom) turned out 37% higher clock because of it. But we also find out that it was uncompetitive in the end that they got kicked out of the market. So it was 37% better than their own process. Where's the 3.5+ year advantage? Did pre-22nm suck so much that even 37% enhancement wasn't enough to make it competitive?

Ivy Bridge? 5% higher clocks.

14nm claimed 40% gain. Where? Oh you mean maybe on the 400MHz IoT chips?

What about density claims? NONE of their products are competitive density wise. It's so far from being competitive that its laughable.

In light of all this, its not surprising at all practically no one else uses Intel's foundry. Their process has a faint chance of shining only when fabbing their own products. It's also probably true that they are very finicky to work with or they charge too much for their services.

 

[witeken]

Ivy Bridge? 5% higher clocks.

14nm claimed 40% gain. Where? Oh you mean maybe on the 400MHz IoT chips?

Lol, why are you hitting on Intel with this?

The first table: http://www.anandtech.com/show/10704/globalfoundries-updates-roadmap-7-nm-in-2h-2018

Further, the issue with product comparison is that it's not apples to apples. Comparing a chip that can go to >4.5GHz versus one that goes to 2.4GHz, for example. Those things have different design tradeoffs, so the comparison will not be process node A vs proces node B. It's like comparing Steamroller to Kaby Lake and concluding that 28nm has higher performance because it clocks higher.