14 nm process means 42 nm half pitch. 10 nm process is proven concept.. good to see it is in place (or going to be... who wins the game will remain to be see). 7 or below, the cost may be difficult and testing are more tricky... new device technology might be better... meso spintronic?
Thanks for finding that link -- I've seen it before, but couldn't put my virtual hand on it.
But my impression is that it's even more complicated than that (see Hattig's post, for example).
Another complication is that there are other variables that affect the ultimate performance of a SOC/CPU/GPU that aren't reflected in that table at all -- factors like TSMC's InFO packaging; IBM's SOI; etc.
In the end trying to compare two different process on the feature size of various elements is very difficult and problematic. What makes the difference in the end is final chip size and thermal usage. Apple in combo with TSMC seems to be hitting some pretty impressive density figures and extremely low power levels.
Beyond all of that it appears that Apples latest A series chips are built on an improved process from TSMC. So the chart referenced above appears to be out of date. An improved process and InFO seems to be a huge win for Apple in the A10 as we are getting a solid jump in performance on what is marketed as a process at the same node. Or to put what you are saying above in different words, the taste and texture of the cake depends upon the cook and the ingredients mixed into the batter.
The question then remains could Intel offer the equivalent performance today as TSMC is with the A10 solution. I think it is fair to say they can't. Effectively TSMC has eclipsed Intel for state of the art products.
Let's not get too excited about this. All that matters for Apple, assuming that they are going to stay with TSMC for a few years, is the improvement of TSMC's processes when compared to older TSMC processes.
intel will remain ahead for the next three or four years, at least. But that doesn't matter unless Apple moves to them.
I don't buy the idea that Intel is ahead anymore. For one A10 is supposedly building on an improved process at TSMC. Beyond that I don't think Intel has the chops to deliver anything close to InFO like TSMC is offering.
To put it another way I don't think Intel can offer Apple an equivalent solution to what TSMC is offering in the A10. Not even close really.
Its better to look at SoC density at a macro level than a micro level when comparing process density between TSMC, Intel and others. A quick comparison of the 16nm (16FFC) A10 vs. the 14nm Quad Skylake shows that TSMC/Apple squeeze in a much higher transistor density. Skylake has higher performance but does less in terms of an SoC (the PCH is a separate chip, plus A10 includes things like a sensor hub as well).
Quad-core + GPU GT2 Core i7 Skylake K 1.75B Transistors 122mm^2 Apple A10 3.3B Transistors 125mm^2
The head of manufacturing has already done analysis about how Apple/TSMC has managed to obtain greater transistor density. - Apple uses a higher percentage of SRAM in their SoCs (though other research indicates that Intel mis-categorized some of Apple's reg. files as SRAM) - Intel uses a higher percentage of cells that are larger than the minimum process geometries for performance (tall cells). - Intel acknowledges that the Apple A8 (TSMC 20nm) and Apple A9 (16nm TSMC and 14nm Samsung) have greater RAW TRANSISTOR DENSITY than similar Broadwell (14nm) or Skylake (14nm) SoCs - Intel has created their own notion, a normalized transistor density, that shows them somewhat ahead, but requires some questionable reverse engineering of exactly what types of logic and memory Apple has included in each block of their SoCs. See foils 19 and 20.
The extra 1.7 billion transistors ought to be telling. It is the total solution though that TSMC is offering that leaves Intel in the dust. InFO is a huge win in offering a compelling and state of the art product for Apple. This is what people often mis, Intel has nothing to compete with and has no ability at all to supply Apple at this time.
14 nm process means 42 nm half pitch. 10 nm process is proven concept.. good to see it is in place (or going to be... who wins the game will remain to be see). 7 or below, the cost may be difficult and testing are more tricky... new device technology might be better... meso spintronic?
As we have seen manufactures are very willing to put a spin on their process nodes. In judging who is ahead here I think it is pretty clear TSMC is. What Apple is capable of putting on a die compared to Intel is very impressive. Combine that with the total solution InFO provides and I think it is pretty clear who is in the lead.
Its better to look at SoC density at a macro level than a micro level when comparing process density between TSMC, Intel and others. A quick comparison of the 16nm (16FFC) A10 vs. the 14nm Quad Skylake shows that TSMC/Apple squeeze in a much higher transistor density. Skylake has higher performance but does less in terms of an SoC (the PCH is a separate chip, plus A10 includes things like a sensor hub as well).
Quad-core + GPU GT2 Core i7 Skylake K 1.75B Transistors 122mm^2 Apple A10 3.3B Transistors 125mm^2
The head of manufacturing has already done analysis about how Apple/TSMC has managed to obtain greater transistor density. - Apple uses a higher percentage of SRAM in their SoCs (though other research indicates that Intel mis-categorized some of Apple's reg. files as SRAM) - Intel uses a higher percentage of cells that are larger than the minimum process geometries for performance (tall cells). - Intel acknowledges that the Apple A8 (TSMC 20nm) and Apple A9 (16nm TSMC and 14nm Samsung) have greater RAW TRANSISTOR DENSITY than similar Broadwell (14nm) or Skylake (14nm) SoCs - Intel has created their own notion, a normalized transistor density, that shows them somewhat ahead, but requires some questionable reverse engineering of exactly what types of logic and memory Apple has included in each block of their SoCs. See foils 19 and 20.
it seems to me that the "normalized transistor density" is closer to being the right thing to look at than just looking at the transistor counts and die sizes without any adjustment at all.
But still.... it's an incomplete picture. The real test would be the same SOC on both processes, but that's not going to happen.
This report at EE Times compares various processes. It concludes that Intel is one-half node ahead of TSMC and is likely to hold that lead over the coming year. But money drives R&D, so TSMC fortified by big orders from Apple has surprised almost every observer in recent months, including Intel:
www.eetimes.com/document.asp?doc_id=1329279
Other elements of the equation include price and responsiveness to customer needs. Shortcomings in these areas, not a lack of technical prowess, explain why Intel has largely been bypassed by the mobile computing revolution -- that is, except what it gained by subsidizing copycat makers of Apple devices. That gamble didn't pay off.
Now that TSMC, Samsung and Globalfoundries processes are within shouting distance of Intel's, if they continue selling (collectively) 1.5 billion units every year, then before long their R&D activities will bring them even with Intel. Moore's Law is driven by costly research.
gatorguy said: I think 'fear of being photographed/recorded' was what led to the downfall of Google Glass wasn't it? I don't know of another reason.
They were fugly and the gestures were ridiculous. I remember a SNL skit that made fun of them. Thought they were exaggerating. Once I saw them in person and someone let me try them, I realized it was accurate. The owner explained that he had more money than sense.
Let's not get too excited about this. All that matters for Apple, assuming that they are going to stay with TSMC for a few years, is the improvement of TSMC's processes when compared to older TSMC processes.
intel will remain ahead for the next three or four years, at least. But that doesn't matter unless Apple moves to them.
The problem for Intel is that its harder and harder to get decent yields for their own designed chips on their process (not a fault with the process itself, more the chip).
By the time Apple hits A12 on TSMC, they could indeed emulate a X86 in a good enough way to switchover all their lower end Mac Laptops to ARM. A 3 core high clocked A12 could hit 14000 on geekbench.
Let's not get too excited about this. All that matters for Apple, assuming that they are going to stay with TSMC for a few years, is the improvement of TSMC's processes when compared to older TSMC processes.
intel will remain ahead for the next three or four years, at least. But that doesn't matter unless Apple moves to them.
Intel is ahead? By what measure? The A9X has better GPU performance than any Intel CPU and performing better than any Intel processor in the same power envelope.
The A10X isn't even out yet. The A10 Fusion is a beast and that's on TSMC's current 16 nm FF process using InFO.
Intel is definitely not leading. They are untouchable on the desktop and server. You'll get no argument from me there. But when it comes to low power devices, Apple and TSMC are substantially better than Intel. It isn't even close.
I can edit 4k video on the iPad Pro. My son's surface book at double the cost chokes on 4k video when in tablet mode. And when in laptop mode, using the discrete GPU, the battery life is substantially inferior to my iPad Pro. It isn't even a contest.
That's all on the A9X. The release of the A10X is imminent.
Intel is not ahead. They had serious troubles with yields getting to their "superior" 14 nm process while TSMC was building very large numbers of A series CPUs for Apple. And TSMC is moving to 10 nm for the manufacturing of the A10X SOC.
I respect your opinions a great deal. They are insightful with thoughtfulness. However, TSMC does not trail Intel by that much. With Info and 16 nm, TSMC's process is actually better than Intel's current 14 nm process.
I realize that Intel's 14 nm process is denser than TSMC's 16 nm. But even with that, Apple is getting performance out of their chips that Intel can't touch at the same TDP.
Over on semiwiki, the processor experts have TSMC seriously ahead of Intel, even predicting that Intel will exit the foundry business.
I do not know where the 3 to 4 year number comes from with respect to Intel leading the other foundries. Intel's foundry business is also more costly and less flexible.
TSMC has some very big plans. And TSMC builds processes specifically to manufacture Apple's SOCs. With the guaranteed revenues from Apple's business, TSMC is spending substantial sums on Capex.
This is a race where Intel can no longer keep up. It is time to face the music. Apple and TSMC are the new powerhouse, displacing Microsoft and Intel.
This report at EE Times compares various processes. It concludes that Intel is one-half node ahead of TSMC and is likely to hold that lead over the coming year. But money drives R&D, so TSMC fortified by big orders from Apple has surprised almost every observer in recent months, including Intel:
www.eetimes.com/document.asp?doc_id=1329279
Other elements of the equation include price and responsiveness to customer needs. Shortcomings in these areas, not a lack of technical prowess, explain why Intel has largely been bypassed by the mobile computing revolution -- that is, except what it gained by subsidizing copycat makers of Apple devices. That gamble didn't pay off.
Now that TSMC, Samsung and Globalfoundries processes are within shouting distance of Intel's, if they continue selling (collectively) 1.5 billion units every year, then before long their R&D activities will bring them even with Intel. Moore's Law is driven by costly research.
Intel is not one half node ahead. Perhaps if lower yields are accounted for.
Apple does not bin the A series chips. They ALL run at the same clock rate. Unlike Intel who separates their better performing CPUs from the lower ones and assigning them as i7s, i5s and i3s. There are likely many A series chips that could perform well at higher speeds. The fact that the whole lot performs at close to Intel's highest end chip line says a lot.
Intel is behind TSMC considering the numbers of manufactured chips each company actually delivers.
This isn't like the old days of AMD where Intel could use their sheer size and budget to overwhelm their rivals.
Intel is being overwhelmed. And had QCOM decided to focus only on modems and not competing directly with Apple in the design and manufacture of Snapdragon CPUs, Intel may not even have gotten Apple's business. As it is, Intel will very likely get all of Apple's modem business one day and deprive QCOM of a very important revenue source. Such a move will keep Intel around for a long time as Apple will be willing to work closely with them for their modems, keeping them competitive or even ahead of the others.
This link succinctly summarizes TSMC and Intel as competitors as pure play foundries. TSMC is ahead and it isn't trivial.
One area where Apple's constant focus on miniaturization may really help in AR/VR headgear: glasses, goggles, etc. One limiting factor on widespread adoption of AR/VR has been the weight -- and frankly the goofy look -- of the headgear. If Apple can shrink that down to a pair of normal-looking sunglasses, it could be a huge tipping point for the industry.
Even if it includes a camera? Serious question. I think 'fear of being photographed/recorded' was what led to the downfall of Google Glass wasn't it? I don't know of another reason.
the honking huge gear on face was a reason. the asymmetrical unease was another.
3nm process?? Incredible. I though it was supposed to max out at 7-10nm. So iPhone 8 chip will be on 10nm process?
A bold move would be for Apple to build the A11X first and deliver it for the 12.7 inch iPad Pro 2 update in early spring, skipping the A10X, then follow up with the A11 in the iPhone 8.
The bold move would be a process shrunk A10 with some minor modifications. The modifications being larger cache, all four cores running at equal performance, and faster memory interfaces. Being process shrunk this A10X could run a lot faster at the same power levels. We might see 30-50 percent over the A10 with little effort on Apples part.
The important thing here is that iPad is low enough volume to do on a process ramp up. By the time TSMC is up to speed Apple will be taping out A11 chips for next years iPhone.
I was actually a bit disappointed that an iPad didn't arrive with the iPhones. However if such a chip could make it into an iPad in the next couple of months it would well be worth the wait. No matter what people say about Apples hardware in general what can't be denied is that the A series chips have become awesome. I may get frustrated with the stupidity of the MiFi program and proprietary connectors but it is hard to complain about iOS device performance theses days. Effectively nobody comes close.
What do you think are the chances that Apple puts one of their custom ARM SoC's into Macs? There seems to be some evidence mounting:
3nm process?? Incredible. I though it was supposed to max out at 7-10nm. So iPhone 8 chip will be on 10nm process?
A bold move would be for Apple to build the A11X first and deliver it for the 12.7 inch iPad Pro 2 update in early spring, skipping the A10X, then follow up with the A11 in the iPhone 8.
The bold move would be a process shrunk A10 with some minor modifications. The modifications being larger cache, all four cores running at equal performance, and faster memory interfaces. Being process shrunk this A10X could run a lot faster at the same power levels. We might see 30-50 percent over the A10 with little effort on Apples part.
The important thing here is that iPad is low enough volume to do on a process ramp up. By the time TSMC is up to speed Apple will be taping out A11 chips for next years iPhone.
I was actually a bit disappointed that an iPad didn't arrive with the iPhones. However if such a chip could make it into an iPad in the next couple of months it would well be worth the wait. No matter what people say about Apples hardware in general what can't be denied is that the A series chips have become awesome. I may get frustrated with the stupidity of the MiFi program and proprietary connectors but it is hard to complain about iOS device performance theses days. Effectively nobody comes close.
What do you think are the chances that Apple puts one of their custom ARM SoC's into Macs? There seems to be some evidence mounting:
Well,what I would like to see and what Apple will do are two different things. Probably the understatement of the year!
As far as putting one in a Mac, Apples ARM chips are already good enough for the Mac Book. However those same chips can be tweaked to provide better performance through larger caches and more memory. I suspect that an A10X type chip will be faster than even Intel's next iteration of core M. Note that is an A10 as expected for an IPad, if Apple generates a 28'watt device the MBA will go ARM too with little resistance. The only place I expect Apple to hold back would be on the MBP's.
In in most cases I believe the media is missing the deeper reasons why Apple would go this route. The cost savings is one element of course but there are further challenges that require that Apple have access to the silicon. As we have seen with the current A series chips Apple puts a lot of tech into the processor design that has nothing to do with the CPU cores. This customization has had a dramatic impact on viability for things like camera processing, sensor handling and so forth. I mentioned it before but silicon is where today's Apple engineer put their IP, innApples Dalit years the engineers put their efforts into interconnected parts on a printed circuit board. Later IP went into PLA and gate arrays. Those days are gone, IP out of necessity has to go onto silicon as part of a SoC build.
Why is access to the silicon required? It is all about physics, you need as much as possible on one chip to avoid, the thermal going off chip introduce and you need the close arrangement of parts due to the time required to propagate signals on the chip. As clock rates increase the length of time it takes to push a signal off chip to another becomes very significant. This is one of the neat things about TSMC InFO technology, it allows physical connection if two dies without the high powered drivers normally required. This is actually huge and is one reason why I see TSMC in the lead technology wise.
Now the the question becomes just how much RAM can be supported with InFO. Will A10X go to 4GB, maybe 8GB. 4 GB means one hell of an iPad upgrade, 8GB means that entry level PCs can be produced. This may sound like a lot of RAM to build into a processor package and it was at one time but today we have even more RAM being built into processor packages with the likes of HBM via AMD and Mamory Cube tech via Intel and Xeon Phi. Different technologies for sure but it does highlight the state of the art. Incidentally one of the reasons these memory technologies are being implemented in package is the same as above short runs enable Apple Has and lower power usage.
In in the end I don't think Apple Has any choice unless they can get Intel or AMD to be more flexible with third party IP and customs chips. Even then i86 is incredibly bloated when all you need is a clean 64 bit architecture these days. Long term i86 is pretty dead technology. Consumers really don't give a damn anymore as everybody has an ARM based tablet or cell phone. The few pros that do will simply run MBP's until they realize just how backwards they are.
So so what do I think, I think I would love to see an ARM based laptop from Apple. Done well I could see a bunch of Linux based ARM machines coming on the market to compete. Now will Apple do this, good question, I'm obviously hopeful! As for the referenced article well it could very well be what they claim but it could also be an indication that they want users to download optimized solutions for all for the currently supported architectures. The tool chains for ARM and i86 are closely linked the entries we see might just be there for completeness.
makes you wonder what an Apple Watch with a 2nm chip would be capable of....
That may not even be feasible considering how close you're getting to the diameter of an atom.
Sometime in the near future we will likely see the return of the clock rate race as process shrinks will be a thing of the past. Clock rate jumps won't be coming from silicon technologies though. There are all,sorts of solutions being worked on in labs. Imagine though an Apple Watch running at 20GHz with a week long battery life.
20GHZ sounds like a lot but even with today's technology we are about a quarter of the way there. An iPad running at 3-4 GhZ is possible in 2017 if Apple wants to go fast. There is good reason to add more cores though even in an iPad. In a laptop you could easily get both. This with today's technology, Apple could easily deliver a six to eight core ARM based laptop chip running at 3.5 GHz. This would crush anything AMD has to offer and a good portion of Intel's line up. The six cores would be easy especially if the IP on board the SoC was optimized for laptop duty not mobile duty.
We we could very well see these later this month,,at least I the MB and possibly the MBA. That might explain the very long delay between machine updates.
Well,what I would like to see and what Apple will do are two different things. Probably the understatement of the year!
As far as putting one in a Mac, Apples ARM chips are already good enough for the Mac Book. ... So so what do I think, I think I would love to see an ARM based laptop from Apple. Done well I could see a bunch of Linux based ARM machines coming on the market to compete. Now will Apple do this, good question, I'm obviously hopeful! As for the referenced article well it could very well be what they claim but it could also be an indication that they want users to download optimized solutions for all for the currently supported architectures. The tool chains for ARM and i86 are closely linked the entries we see might just be there for completeness.
First, let me say that all this technical discussion on the chips is why I come to these forums, this thread is filled with fantastic information and discussion, thanks for sharing and participating!
My question has to do with ARM on laptops, it's something I think none would be surprised to see from Apple at some point, but what I never see is any discussion around the merits of not running macOS on an ARM laptop, but instead a variant of iOS. iOS has a much larger install base, and I would assume a much larger developer community. There is already a non-touch version of iOS in Apple ARM devices (tvOS), so I would assume turning the OS into a proper non-touch OS such that it could be run on a laptop is probably not a lot of work on Apple's part. From a business perspective doesn't this make more sense than porting macOS to ARM and losing what everyone always decries when it's suggested (macOS on ARM) with their "what about Windows?!" concerns, plus also the porting of all existing apps is not a trivial affair as we've already been through a decade ago. How many people would jump ship from Apple laptop/desktops, and would that be offset by new entrants to their user base with macOS on ARM?
I guess Apple would be asking themselves how many more laptops (and desktops) could they sell with macOS on ARM, and would that be more or less than those same laptops and desktops running a non-touch iOS? It seems to me that people who don't normally buy laptops or desktops from Apple but have one or more iDevices already would be prime candidates for these new machines, and that's a HUGE base from which to draw future sales.
Keeping Intel chips on some laptops and desktops would allow them to continue selling more expensive hardware to niche users, those with higher computing needs and special use cases, and not piss off the current install base or developer community or Intel itself, plus it would allow them to pursue an entirely new path of users with ARM chips in new devices, expanding that base and developer community (this seems a very important factor!). Doesn't this seem a more logical path to take?
Putting macOS on ARM another point people bring up immediately is, "macOS on an iPad." This is a fair question, but what makes more sense: macOS on iPad or (non-touch) iOS on laptop/desktop? iOS has matured so much in the past few years, they keep pushing this as more than a "toy OS" and there is less and less that separates it from a desktop/laptop OS each iteration, creating another variant of the OS to run in a non-touch environment just seems a logical extension, especially when one considers they've already done it at least partially and it's been in market for a year now. Which would you prefer? I know which one excites me more (both are interesting of course). Which do you think is more likely?
Well,what I would like to see and what Apple will do are two different things. Probably the understatement of the year!
As far as putting one in a Mac, Apples ARM chips are already good enough for the Mac Book. ... So so what do I think, I think I would love to see an ARM based laptop from Apple. Done well I could see a bunch of Linux based ARM machines coming on the market to compete. Now will Apple do this, good question, I'm obviously hopeful! As for the referenced article well it could very well be what they claim but it could also be an indication that they want users to download optimized solutions for all for the currently supported architectures. The tool chains for ARM and i86 are closely linked the entries we see might just be there for completeness.
First, let me say that all this technical discussion on the chips is why I come to these forums, this thread is filled with fantastic information and discussion, thanks for sharing and participating!
My question has to do with ARM on laptops, it's something I think none would be surprised to see from Apple at some point, but what I never see is any discussion around the merits of not running macOS on an ARM laptop, but instead a variant of iOS. iOS has a much larger install base, and I would assume a much larger developer community. There is already a non-touch version of iOS in Apple ARM devices (tvOS), so I would assume turning the OS into a proper non-touch OS such that it could be run on a laptop is probably not a lot of work on Apple's part. From a business perspective doesn't this make more sense than porting macOS to ARM and losing what everyone always decries when it's suggested (macOS on ARM) with their "what about Windows?!" concerns, plus also the porting of all existing apps is not a trivial affair as we've already been through a decade ago. How many people would jump ship from Apple laptop/desktops, and would that be offset by new entrants to their user base with macOS on ARM?
I guess Apple would be asking themselves how many more laptops (and desktops) could they sell with macOS on ARM, and would that be more or less than those same laptops and desktops running a non-touch iOS? It seems to me that people who don't normally buy laptops or desktops from Apple but have one or more iDevices already would be prime candidates for these new machines, and that's a HUGE base from which to draw future sales.
Keeping Intel chips on some laptops and desktops would allow them to continue selling more expensive hardware to niche users, those with higher computing needs and special use cases, and not piss off the current install base or developer community or Intel itself, plus it would allow them to pursue an entirely new path of users with ARM chips in new devices, expanding that base and developer community (this seems a very important factor!). Doesn't this seem a more logical path to take?
Putting macOS on ARM another point people bring up immediately is, "macOS on an iPad." This is a fair question, but what makes more sense: macOS on iPad or (non-touch) iOS on laptop/desktop? iOS has matured so much in the past few years, they keep pushing this as more than a "toy OS" and there is less and less that separates it from a desktop/laptop OS each iteration, creating another variant of the OS to run in a non-touch environment just seems a logical extension, especially when one considers they've already done it at least partially and it's been in market for a year now. Which would you prefer? I know which one excites me more (both are interesting of course). Which do you think is more likely?
Excellent reply & discuss by both you and wizard69! Regarding your suggestion of Apple doing a non-touch version of iOS on a laptop / desktop - if they did that, they wouldn't be that far away from doing a "2-in-1" (as opposed to having a separate iPad & laptop line). The only difference being that Apple's version, being based off of iOS, wouldn't run desktop apps but that wouldn't matter considering the target audience. There's 2 ways Apple could go about it:
1. Adding KB / mouse capability to iOS would allow the iPad Pro to dynamically optimize the UI (touch or non-touch) based on if you have a KB attached (or not). This would be similar to MS' Surface Pro Continuum feature
it seems to me that the "normalized transistor density" is closer to being the right thing to look at than just looking at the transistor counts and die sizes without any adjustment at all.
But still.... it's an incomplete picture. The real test would be the same SOC on both processes, but that's not going to happen.
transistor density is great for apple. hopefully, the consumption of energy and delay are posted as well... or real life test would be nice. I do like the A10 move ahead... it provide apple unique advantage in the headset competitive field. just for the density alone, I can't tell which one is better, intel or apple....In addition, as with all the ultra-fast processor, off chip interconnect delay would impact performance more than before... apple, has in house off chip... might have advantage there... Intel got great product, however they users (or their customer) are not all the masters of design... many just copy cat (especially, those don't have that much IP in house and steal others for living).
Comments
Beyond all of that it appears that Apples latest A series chips are built on an improved process from TSMC. So the chart referenced above appears to be out of date. An improved process and InFO seems to be a huge win for Apple in the A10 as we are getting a solid jump in performance on what is marketed as a process at the same node. Or to put what you are saying above in different words, the taste and texture of the cake depends upon the cook and the ingredients mixed into the batter.
The question then remains could Intel offer the equivalent performance today as TSMC is with the A10 solution. I think it is fair to say they can't. Effectively TSMC has eclipsed Intel for state of the art products.
But still.... it's an incomplete picture. The real test would be the same SOC on both processes, but that's not going to happen.
www.eetimes.com/document.asp?doc_id=1329279
Other elements of the equation include price and responsiveness to customer needs. Shortcomings in these areas, not a lack of technical prowess, explain why Intel has largely been bypassed by the mobile computing revolution -- that is, except what it gained by subsidizing copycat makers of Apple devices. That gamble didn't pay off.
Now that TSMC, Samsung and Globalfoundries processes are within shouting distance of Intel's, if they continue selling (collectively) 1.5 billion units every year, then before long their R&D activities will bring them even with Intel. Moore's Law is driven by costly research.
By the time Apple hits A12 on TSMC, they could indeed emulate a X86 in a good enough way to switchover all their lower end Mac Laptops to ARM.
A 3 core high clocked A12 could hit 14000 on geekbench.
The A10X isn't even out yet. The A10 Fusion is a beast and that's on TSMC's current 16 nm FF process using InFO.
Intel is definitely not leading. They are untouchable on the desktop and server. You'll get no argument from me there. But when it comes to low power devices, Apple and TSMC are substantially better than Intel. It isn't even close.
I can edit 4k video on the iPad Pro. My son's surface book at double the cost chokes on 4k video when in tablet mode. And when in laptop mode, using the discrete GPU, the battery life is substantially inferior to my iPad Pro. It isn't even a contest.
That's all on the A9X. The release of the A10X is imminent.
Intel is not ahead. They had serious troubles with yields getting to their "superior" 14 nm process while TSMC was building very large numbers of A series CPUs for Apple. And TSMC is moving to 10 nm for the manufacturing of the A10X SOC.
I respect your opinions a great deal. They are insightful with thoughtfulness. However, TSMC does not trail Intel by that much. With Info and 16 nm, TSMC's process is actually better than Intel's current 14 nm process.
I realize that Intel's 14 nm process is denser than TSMC's 16 nm. But even with that, Apple is getting performance out of their chips that Intel can't touch at the same TDP.
Over on semiwiki, the processor experts have TSMC seriously ahead of Intel, even predicting that Intel will exit the foundry business.
I do not know where the 3 to 4 year number comes from with respect to Intel leading the other foundries. Intel's foundry business is also more costly and less flexible.
TSMC has some very big plans. And TSMC builds processes specifically to manufacture Apple's SOCs. With the guaranteed revenues from Apple's business, TSMC is spending substantial sums on Capex.
This is a race where Intel can no longer keep up. It is time to face the music. Apple and TSMC are the new powerhouse, displacing Microsoft and Intel.
Apple does not bin the A series chips. They ALL run at the same clock rate. Unlike Intel who separates their better performing CPUs from the lower ones and assigning them as i7s, i5s and i3s. There are likely many A series chips that could perform well at higher speeds. The fact that the whole lot performs at close to Intel's highest end chip line says a lot.
Intel is behind TSMC considering the numbers of manufactured chips each company actually delivers.
This isn't like the old days of AMD where Intel could use their sheer size and budget to overwhelm their rivals.
Intel is being overwhelmed. And had QCOM decided to focus only on modems and not competing directly with Apple in the design and manufacture of Snapdragon CPUs, Intel may not even have gotten Apple's business. As it is, Intel will very likely get all of Apple's modem business one day and deprive QCOM of a very important revenue source. Such a move will keep Intel around for a long time as Apple will be willing to work closely with them for their modems, keeping them competitive or even ahead of the others.
This link succinctly summarizes TSMC and Intel as competitors as pure play foundries. TSMC is ahead and it isn't trivial.
http://www.barrons.com/articles/BL-TB-52894
http://www.idownloadblog.com/2016/09/30/macos-sierra-code-suggests-apple-could-replace-intel-in-macs-with-custom-arm-chips/
As far as putting one in a Mac, Apples ARM chips are already good enough for the Mac Book. However those same chips can be tweaked to provide better performance through larger caches and more memory. I suspect that an A10X type chip will be faster than even Intel's next iteration of core M. Note that is an A10 as expected for an IPad, if Apple generates a 28'watt device the MBA will go ARM too with little resistance. The only place I expect Apple to hold back would be on the MBP's.
In in most cases I believe the media is missing the deeper reasons why Apple would go this route. The cost savings is one element of course but there are further challenges that require that Apple have access to the silicon. As we have seen with the current A series chips Apple puts a lot of tech into the processor design that has nothing to do with the CPU cores. This customization has had a dramatic impact on viability for things like camera processing, sensor handling and so forth. I mentioned it before but silicon is where today's Apple engineer put their IP, innApples Dalit years the engineers put their efforts into interconnected parts on a printed circuit board. Later IP went into PLA and gate arrays. Those days are gone, IP out of necessity has to go onto silicon as part of a SoC build.
Why is access to the silicon required? It is all about physics, you need as much as possible on one chip to avoid, the thermal going off chip introduce and you need the close arrangement of parts due to the time required to propagate signals on the chip. As clock rates increase the length of time it takes to push a signal off chip to another becomes very significant. This is one of the neat things about TSMC InFO technology, it allows physical connection if two dies without the high powered drivers normally required. This is actually huge and is one reason why I see TSMC in the lead technology wise.
Now the the question becomes just how much RAM can be supported with InFO. Will A10X go to 4GB, maybe 8GB. 4 GB means one hell of an iPad upgrade, 8GB means that entry level PCs can be produced. This may sound like a lot of RAM to build into a processor package and it was at one time but today we have even more RAM being built into processor packages with the likes of HBM via AMD and Mamory Cube tech via Intel and Xeon Phi. Different technologies for sure but it does highlight the state of the art. Incidentally one of the reasons these memory technologies are being implemented in package is the same as above short runs enable Apple Has and lower power usage.
In in the end I don't think Apple Has any choice unless they can get Intel or AMD to be more flexible with third party IP and customs chips. Even then i86 is incredibly bloated when all you need is a clean 64 bit architecture these days. Long term i86 is pretty dead technology. Consumers really don't give a damn anymore as everybody has an ARM based tablet or cell phone. The few pros that do will simply run MBP's until they realize just how backwards they are.
So so what do I think, I think I would love to see an ARM based laptop from Apple. Done well I could see a bunch of Linux based ARM machines coming on the market to compete. Now will Apple do this, good question, I'm obviously hopeful! As for the referenced article well it could very well be what they claim but it could also be an indication that they want users to download optimized solutions for all for the currently supported architectures. The tool chains for ARM and i86 are closely linked the entries we see might just be there for completeness.
Sometime in the near future we will likely see the return of the clock rate race as process shrinks will be a thing of the past. Clock rate jumps won't be coming from silicon technologies though. There are all,sorts of solutions being worked on in labs. Imagine though an Apple Watch running at 20GHz with a week long battery life.
20GHZ sounds like a lot but even with today's technology we are about a quarter of the way there. An iPad running at 3-4 GhZ is possible in 2017 if Apple wants to go fast. There is good reason to add more cores though even in an iPad. In a laptop you could easily get both. This with today's technology, Apple could easily deliver a six to eight core ARM based laptop chip running at 3.5 GHz. This would crush anything AMD has to offer and a good portion of Intel's line up. The six cores would be easy especially if the IP on board the SoC was optimized for laptop duty not mobile duty.
We we could very well see these later this month,,at least I the MB and possibly the MBA. That might explain the very long delay between machine updates.
My question has to do with ARM on laptops, it's something I think none would be surprised to see from Apple at some point, but what I never see is any discussion around the merits of not running macOS on an ARM laptop, but instead a variant of iOS. iOS has a much larger install base, and I would assume a much larger developer community. There is already a non-touch version of iOS in Apple ARM devices (tvOS), so I would assume turning the OS into a proper non-touch OS such that it could be run on a laptop is probably not a lot of work on Apple's part. From a business perspective doesn't this make more sense than porting macOS to ARM and losing what everyone always decries when it's suggested (macOS on ARM) with their "what about Windows?!" concerns, plus also the porting of all existing apps is not a trivial affair as we've already been through a decade ago. How many people would jump ship from Apple laptop/desktops, and would that be offset by new entrants to their user base with macOS on ARM?
I guess Apple would be asking themselves how many more laptops (and desktops) could they sell with macOS on ARM, and would that be more or less than those same laptops and desktops running a non-touch iOS? It seems to me that people who don't normally buy laptops or desktops from Apple but have one or more iDevices already would be prime candidates for these new machines, and that's a HUGE base from which to draw future sales.
Keeping Intel chips on some laptops and desktops would allow them to continue selling more expensive hardware to niche users, those with higher computing needs and special use cases, and not piss off the current install base or developer community or Intel itself, plus it would allow them to pursue an entirely new path of users with ARM chips in new devices, expanding that base and developer community (this seems a very important factor!). Doesn't this seem a more logical path to take?
Putting macOS on ARM another point people bring up immediately is, "macOS on an iPad." This is a fair question, but what makes more sense: macOS on iPad or (non-touch) iOS on laptop/desktop? iOS has matured so much in the past few years, they keep pushing this as more than a "toy OS" and there is less and less that separates it from a desktop/laptop OS each iteration, creating another variant of the OS to run in a non-touch environment just seems a logical extension, especially when one considers they've already done it at least partially and it's been in market for a year now. Which would you prefer? I know which one excites me more (both are interesting of course). Which do you think is more likely?
1. Adding KB / mouse capability to iOS would allow the iPad Pro to dynamically optimize the UI (touch or non-touch) based on if you have a KB attached (or not). This would be similar to MS' Surface Pro Continuum feature
2. Second way is doing something similar to this patent here: http://www.patentlyapple.com/patently-apple/2013/04/finally-apple-reveals-their-hybrid-notebook-tablet-details.html
Just my 2 cents. Again, excellent discussion all around.
But still.... it's an incomplete picture. The real test would be the same SOC on both processes, but that's not going to happen. transistor density is great for apple. hopefully, the consumption of energy and delay are posted as well... or real life test would be nice. I do like the A10 move ahead... it provide apple unique advantage in the headset competitive field. just for the density alone, I can't tell which one is better, intel or apple....In addition, as with all the ultra-fast processor, off chip interconnect delay would impact performance more than before... apple, has in house off chip... might have advantage there... Intel got great product, however they users (or their customer) are not all the masters of design... many just copy cat (especially, those don't have that much IP in house and steal others for living).