(Theory but possible) Efficient cores might be the future of multi-core performance

[dmccloud]

Do you think they'll get it under control, or keep up the factory overclocks to chase a number?

It would be nice to build a desktop that doesn't require adding a second boiler to the home powerplant.

I'm not sure Intel can engineer any other way but overclocks at this stage. "Moore's Law" has morphed from being a driving force pushing Intel forward to a metaphorical wall they keep banging their heads against and devising ways to get around said wall. A lot of Intel's talent that drove the company forward in chip design and fabrication have left the company, and those left behind simply have not been able to keep up the pace of prior regimes at Intel, which is a major reason there has been so much turnover at the executive level there over the last 18 months.

 

[Toutou]

8 performance cores + 120 efficient cores

This is how we build GPUs, a crazy amount of tiny, simple cores. We do it because the kind of work that a GPU does is easily parallelizable, it's mostly the same sequence of operations over many objects at once.

On the other hand, the kind of code that we usually run on the CPU is usually much more complicated (like orders of magnitude more complicated) and is often much more difficult (or, in some cases, impossible) to parallelize. And even for the tasks that are easily split into multiple "jobs" (they're called "threads" in code and in hardware) there quickly comes the area of diminishing returns where adding more threads only burns more electricity and overhead (synchronization, memory access) eats most of the added performance.

So there's actually a sweet spot for user oriented CPUs where it's still possible for the single user sitting in front of the machine to utilize all the cores without wasting too much electricity and without the most important threads running too slow because everyone else is thrashing the RAM and I/O. And that sweet spot, as far as we know, is still much lower than 128 cores. And as for the kind of workloads that scale really well over 128 cores (scientific stuff, AI, video encoding), we're (slowly) trying to move those over to our GPUs, and we're actually making those capable of running more complex code.

 

[jdb8167]

As I stated earlier, this only affects 3nm processes. It's not like Apple faces a significant disadvantage on the CPU/SoC front with the 5nm process, especially given the scalability and modularity inherent to Apple's silicon designs. Apple could probably stay on 5nm/5nm+ for now and just jump to 2nm when it's ready, which would again put Intel behind the rest of the industry. AMD is also a major customer of TSMC, so perhaps Apple and AMD work together to outmaneuver Intel on the move to 2nm. What doesn't pass the logic test is trying to address Intel's move to lock up all remaining 3nm production in a vacuum without taking into account the options available to both Apple and AMD regarding TSMC capabilities and processes.

Well that techradar article contradicts the earlier windowscentral article. Which one are you going to believe? The idea that TSMC’s largest customer got blocked out by Intel on 3nm fabrication doesn’t pass the most basic test of logic. Apple has the best supply chain logistics of any company in the world yet we are supposed to believe that they made a mistake and didn’t lock down the next generation fabrication. Not buying it.

 

[leman]

Do you think they'll get it under control, or keep up the factory overclocks to chase a number?

It doesn’t seem like they have a way to get it under control to be honest. The latest strategy is to make the performance cores even more power hungry and compensate by using efficiently cores where multi-core performance is required. The announced performance crease of the new Adler Lake performance core is impressive, but there is little doubt that it is achieved at the expense of power consumption.

To put it in perspective, it means that Apple has much more wiggle room. They currently need 5 watts to deliver performance where Intel needs 20 watts or more. So even if Apple is stuck n extracting more performance out of their designs, they still always have the option to increase the power consumption. If their next-gen core draws 10 watts to improve performance by mere 25%, they still have Intel beat while consuming half the power.

 

[jdb8167]

It doesn’t seem like they have a way to get it under control to be honest. The latest strategy is to make the performance cores even more power hungry and compensate by using efficiently cores where multi-core performance is required. The announced performance crease of the new Adler Lake performance core is impressive, but there is little doubt that it is achieved at the expense of power consumption.

To put it in perspective, it means that Apple has much more wiggle room. They currently need 5 watts to deliver performance where Intel needs 20 watts or more. So even if Apple is stuck n extracting more performance out of their designs, they still always have the option to increase the power consumption. If their next-gen core draws 10 watts to improve performance by mere 25%, they still have Intel beat while consuming half the power.

These leaked TDP numbers are telling. I would assume that the top performance core single-core results are coming from the supposed 45 W Alder Lake-P which can reach 215 W.

Edit: Actually those other results are from an 8+8 part which means desktop. These TDP numbers are for mobile. Still makes the point though about relative power vs the M1.

Alder Lake Specs. The nomenclature is x+y+z where x is the # of performance cores, y efficiency cores, and z graphics level which are all GT2 level with 96 execution units.

	Alder Lake-P	Alder Lake-M
	(2 + 8 + 2): <= 15W
PL1 (TDP)	(4 + 8 + 2): <= 28W	(2 + 8 + 2): <= 9W
	(6 + 8 + 2): <= 45W
	(2 + 8 + 2): <= 55W
PL2	(4 + 8 + 2): <= 64W	(2 + 8 + 2): <= 30W
	(6 + 8 + 2): <= 115W
	(2 + 8 + 2): <= 123W
PL4	(4 + 8 + 2): <= 140W	(2 + 8 + 2): <= 68W
	(6 + 8 + 2): <= 215W

 

[leman]

These leaked TDP numbers are telling. I would assume that the top performance core single-core results are coming from the supposed 45 W Alder Lake-P which can reach 215 W.

Alder Lake Specs. The nomenclature is x+y+z where x is the # of performance cores, y efficiency cores, and z graphics level which are all GT2 level with 96 execution units.

	Alder Lake-P	Alder Lake-M
	(2 + 8 + 2): <= 15W
PL1 (TDP)	(4 + 8 + 2): <= 28W	(2 + 8 + 2): <= 9W
	(6 + 8 + 2): <= 45W
	(2 + 8 + 2): <= 55W
PL2	(4 + 8 + 2): <= 64W	(2 + 8 + 2): <= 30W
	(6 + 8 + 2): <= 115W
	(2 + 8 + 2): <= 123W
PL4	(4 + 8 + 2): <= 140W	(2 + 8 + 2): <= 68W
	(6 + 8 + 2): <= 215W

70W peak power draw for a 9W SKU... in which world is this considered normal?

 

[jdb8167]

70W peak power draw for a 9W SKU... in which world is this considered normal?

The 68 W is for PL4 which probably can only be active for milliseconds but the PL2 is turbo and that is 30W. Still ridiculous.

 

[leman]

The 68 W is for PL4 which probably can only be active for milliseconds but the PL2 is turbo and that is 30W. Still ridiculous.

Even if it’s milliseconds, the mere fact that a chassis nominally designed for 9watts will briefly experience power draw exceeding that limit by almost a factor of 10 is insane. That’s can’t be good for the electrical system and of course it will lead to even more vicious spec wars where manufacturers will put a 9W SKU into a 20W chassis to advertise “superior performance“.

 

[JouniS]

So there's actually a sweet spot for user oriented CPUs where it's still possible for the single user sitting in front of the machine to utilize all the cores without wasting too much electricity and without the most important threads running too slow because everyone else is thrashing the RAM and I/O. And that sweet spot, as far as we know, is still much lower than 128 cores. And as for the kind of workloads that scale really well over 128 cores (scientific stuff, AI, video encoding), we're (slowly) trying to move those over to our GPUs, and we're actually making those capable of running more complex code.

GPUs are only good for SIMD-type tasks. You have an embarrassingly parallel problem that can be solved independently for each data point, and the execution paths for adjacent data points are usually the same. Once the execution paths start diverging, GPUs become inefficient.

Five years ago, there was still plenty of enthusiasm for figuring out what can be done efficiently with GPUs. I don't see that enthusiasm anymore. We already know what GPUs are good for, and the rest is a matter of engineering. If you need to execute many independent jobs, CPUs are the right choice. With current memory and I/O architectures, you can usually run tens of parallel jobs efficiently. In some tasks, even hundreds of CPU cores in the same system make sense. In other tasks, a few jobs may already be enough to saturate the bus.

 

[deconstruct60]

As I stated earlier, this only affects 3nm processes.

Extremely likely this isn't the 3nm processes. More likely this is the 3nm process in a specific time window. A quarter or two. The 3nm process is likely going to be produced at TSMC for 3-6 years. No practical way Intel is paying upfront for stuff they will be buying 4-6 years from now. It doesn't work that way. There is ZERO rational reason for TSMC to sell all of the wafers several years out either without minimal payment guarantees from Intel that don't make rational sense for Intel . There is no win/win for both sides there.

What the article is more likely referring to is the initial "at risk" and the first quarter (maybe two ) of high volume production. Intel could be buying large , significant fractions of that.

First, the Ponte Vecchio Xe-HPC compute engine uses TSMC N5. Intel is probably going to want to push the performance of their "super computing" GPU forward aggressively so taking that N3 in another 18 months would make lots of sense. Alternatively, the RAMBO cache is old Intel 10nm. That too probably could benefit from a shift to N3 and help the overall Xe-HPC module get better performance. If Intel pumps out 10,000 Next Gen Xe-HPC chips that is relatively "high volume" for that product. For Apple it is "lost change in the couch don't care about" volumes. It could have some roll out impact on Apple is their 3nm stuff needs to be several early steeping tweaks before going to high volume production.

When TSMC is at 4,000 wafer per month Intel could take large slice. How long it takes to get to 10,000 per month is rather vague in the articles. Which leads to the time problem. Years down the road TSMC doesn't have the same mix of customers on a node they do at the early "at risk" stages.

Second, Apple tends to pragmatically pay fort he ASML scanners they are using at TSMC. This could also be Intel buying up the non-Apple 3nm scanners volume. That wouldn't necessarily put a squeeze on Apple in that case ( well it does if Apple wants to buy more than they thought they would need. And probably certainly drives up the prices Apple is paying; as nobody else is help paying for the rest of the equipment and floor space. )

It's not like Apple faces a significant disadvantage on the CPU/SoC front with the 5nm process, especially given the scalability and modularity inherent to Apple's silicon designs. Apple could probably stay on 5nm/5nm+ for now and just jump to 2nm when it's ready, which would again put Intel behind the rest of the industry. AMD is also a major customer of TSMC, so perhaps Apple and AMD work together to outmaneuver Intel on the move to 2nm.

Very likely there will be a 3P , 3+ , or some other variant tweak. All the previous TSMC nodes had optimized updates that come later. There won't be a need to wait for TSMC N2 .
Which is the other possible path. That Intel (or Apple) is getting a 3-tweaked. The articles could be chasing a single variant which go lost in the "telephone game" translation as the information leaks out of TSMC.

What doesn't pass the logic test is trying to address Intel's move to lock up all remaining 3nm production in a vacuum without taking into account the options available to both Apple and AMD regarding TSMC capabilities and processes.

Intel probably isn't just locking up capacity. The rumors/leaked arrow lake with over a hundred Xe GPU units. If that is a TSMC N3 tile hooked to a Intel N7 CPU core tile could be a high volume SoC that put some dent into any M-series migration drift or M-series envy in the laptop space.

 

[deconstruct60]

The problem there is that Intel jumped ahead of Apple and others and apparently locked up 3nm capacity from TSMC. Granted, that's not necessarily for consumer products, but that still throws a wrench in Apple's move to 3nm. TSMC has been working on 2nm as well, so maybe Apple leapfrogs Intel and jumps from 2nm to 3nm.

Intel to utilize TSMC 3nm technology as it struggles to move to 7nm process, says report

Intel will reportedly utilize TSMC's 3nm technology while it works to improve its own processors.

www.windowscentral.com

" ...
a source told Nikkei Asia that Intel may order more of the new processors, "Currently the chip volume planned for Intel is more than that for Apple's iPad using the 3-nanometer process."

.... "

That is not the same thing as buying up all the wafer starts. There are two states. There are "maybe i'll buy xxx wafers in a couple of years" wafer starts. This helps TSMC plan how big the future plants need to be. It is also usually relatively "free" ( relatively low cost. ) The other is buying a guarantee wafer start ( you are required to process these wafers on these days during these specific months for $ABC money on this specific date. ) . This typically costs money to hold a place in the wafer start line.

This points to Intel providing TSMC the first rather than the latter. Apple probably isn't telling TSMC exact specific requests 2-4 years in advance for the same reason they don't tell customers 1-2 years in advance they are shipping some new product. TSMC will get told when they absolutely need to know and not before.

Intel has lots of competitive upsides at the moment for churning the waters on projected 3nm production. with non-guarantee buy wafer starts. If they need them fine (the capacity should largely be there for them). If they don't then some of their competitors are having to plan around maybe running into shortages. It just keeps the competitors more off balance while Intel regains their footing.

It is probably a bit annoying to TSMC, but they are playing along. Fabs running at 100% capacity makes them more money. Intel probably is going to take profits from their more competitive chips and plow a large chunk of the profits into getting Intel Fabs back into the game. ( so TSMC is facilitating more competition over the longer term. )

If Intel turns out not to buy as much, there is very good chance TSMC will find other customers to fill the void left by Intel ( similar to how TSMC found users for all the lightly earmarked wafer starts for Huawei that got handed back when Huawei got blocked. )

 

[Kung gu]

Intel probably isn't just locking up capacity. The rumors/leaked arrow lake with over a hundred Xe GPU units. If that is a TSMC N3 tile hooked to a Intel N7 CPU core tile could be a high volume SoC that put some dent into any M-series migration drift or M-series envy in the laptop space.

A new report says otherwise and says Apple secure 3nm for 2022.
https://www.gizmochina.com/2021/08/16/apple-secure-majority-tsmc-3nm-over-intel/#:~:text=Apple secures majority of TSMC's 3nm production capacity over Intel: Report,-By&text=While TSMC is on track,has been secured by Apple.

People who buy laptops for gaming will never get a Mac but people who only code, do software engineering and have creative/content creation and audio/music production needs will consider a Mac.

You have to remember Macbook sales were decent when Apple had horrible themals, bad cpus and gpus and bad keyboards.

Now all those are going to be fixed later this year for the high end MacBooks. MacBook sales will go up. This will not be some silent site refresh like in the past for the Macbook pro like how the 16" Intel Mac was announced. Apple will have an event for these M1X Macs. Apple will have all of November, December and January before Alder Lake even launches. Even then Alder Lake won't be as efficient as M1X.

For the people who buy ultrabooks by the time Intel launches the successor to Alder Lake, M2 will come out.
Intel has serious competition now and its not just from Apple. Qualcomm with Nuvia will also be annoying for Intel.

Of course time will tell but interesting times ahead for the laptop/desktop industry.

 

[jdb8167]

A new report says otherwise and says Apple secure 3nm for 2022.
https://www.gizmochina.com/2021/08/16/apple-secure-majority-tsmc-3nm-over-intel/#:~:text=Apple secures majority of TSMC's 3nm production capacity over Intel: Report,-By&text=While TSMC is on track,has been secured by Apple.

Unfortunately this report came from DigiTimes which is just about as unreliable as Nikkei. On the hand, this report makes a lot more logical sense rather than believing that TSMC would short their best customer in favor of Intel who is likely to use TSMC only until their process is back to industry standards with EUV.

 

[deconstruct60]

A new report says otherwise and says Apple secure 3nm for 2022.
https://www.gizmochina.com/2021/08/16/apple-secure-majority-tsmc-3nm-over-intel/#:~:text=Apple secures majority of TSMC's 3nm production capacity over Intel: Report,-By&text=While TSMC is on track,has been secured by Apple.

People who buy laptops for gaming will never get a Mac but people who only code, do software engineering and have creative/content creation and audio/music production needs will consider a Mac.

I suspect that the two reports are grasping different parts of the elephant (one grasps the trunk and calls it a 'snake' and the other grasps the leg and calls it a 'tree' ). The relatively low volume HPC GPU component and very high priced server 3nm parts are "most of the volume" when volume is low and not most of the volume when 3nm is running at full steam high volume production.

It doesn't make sense for TSMC to stiff any of the high end customers they have. Apple isn't that special ( that is why Apple is paying lots of money upfront. Paying upfront doesn't make them special. Just means they paid early. Anyone else with a large enough stack of money probably can get most of what Apple is getting. )

The 2023 Intel grabbing a bigger chunk for volume production for ArrowLake in 2023 lines up with the ArrowLake timeline.

The core problem issue is that neither article reflect the whole lifecycle of the N3 process very well.

If there are 4-5 "big" customers ramping up for major volume products on N3, then Apple will we walking a thinner road on the ramp. It is probably not too thin but mistakes by Apple will likely not be amenable to just 'buying" their way out of . They are more likely going to cost time ( i.e., time to market ).

You have to remember Macbook sales were decent when Apple had horrible themals, bad cpus and gpus and bad keyboards.

Now all those are going to be fixed later this year for the high end MacBooks. MacBook sales will go up. This will not be some silent site refresh like in the past for the Macbook pro like how the 16" Intel Mac was announced. Apple will have an event for these M1X Macs. Apple will have all of November, December and January before Alder Lake even launches. Even then Alder Lake won't be as efficient as M1X.

A couple of issues.

1. a one time "dog and pony" show isn't going to fundamentally change sales one way or the other over the long term. More likely it will cause a supply/demand mismatch that time shifts a minirity amount of sales to "sooner" rather than "later". There will be a "fear of missing out" (FOMO) and several opportunites for scalpers to make a buck... but more sales over a 12-18 month period ? Probably not.

If it is a good product it will sell well. That is far more important than the "dog and pony" show.


2. You're presuming that Apple doesn't toss away thermal cooling capacity by making the updated enclosure thinner ( e.g., the iMac 24" ). It wasn't so much that CPUs they had were "bad" as much as Apple designed the thermals of the enclosure to be "just enough" big enough for the nominal state of the CPU. Not designing for upper edge cases or giving themselves some 'slack' to deal with the "unexpected".

Applying that "minimally large enough" thermals to a M1X will likely lead to similar "bad" thermals just at different levels.

3. Windows laptops will continue to run Windows natively with or without Alder Lake mobile models. ( the desktop variants will show up first in 2021 ). Many of them will have removable RAM DIMMs and SSD drives. They have a selection of USB ports that people use daily.
Apple is suppose to be bringing back HDMI , power socket, and SD card ports, but only at price points the vast majority of the overall PC laptop market doesn't want to pay.

Apple will probably do incrementally better but probably wont "take over" the general PC laptop market. Price will be an issue and frankly they won't be able order that many chips if they even wanted to.

For the people who buy ultrabooks by the time Intel launches the successor to Alder Lake, M2 will come out.
Intel has serious competition now and its not just from Apple. Qualcomm with Nuvia will also be annoying for Intel.

And M2 still won't natively boot Windows (what most people have ) . Or be IT repair friendly . or .....

To a large extent Intel's solutions don't have to be overall faster,. They primarily need to be at least "good enough" (primarily as an upgrade to what folks currently have) and priced to match the market. Intel's margins will likely go down before they start to loose a critical amount of market share. Are they going to loose some share? Yes. Most of it? If the Windows 11 optimizations work well and no hiccups on production ... then probably not.

Qualcomm/Nuvia will be coming on Windows 11 also. It won't be surprising to see the Samsung/AMD mash up become a player in Windows 11 in late 2022 or early 2023 also. And AMD is going to be around also.

But that is a dual edge sword that doesn't necessarily go all Apple's way. Windows 11 is going to have a diverse ecosystem of suppliers. macOS won't.

Of course time will tell but interesting times ahead for the laptop/desktop industry.

The pandemic triggers a large uptick in system purchases over last 18 months. At some point that huge demand surge is going to be followed by a demand drop (from average levels). "Growling" into a falling market will be harder in late 2022-2024. How that intersects with the changes with the base technology coming will be interesting.

It isn't just hardware that is changing. Software wise also. If the Windows 11 more rigorous standards get that platform substantively more stable user experience that will make them more competitive. ( as opposed to the quirky stuff that Vista and Win8 introduced. )

 

[Kung gu]

You're presuming that Apple doesn't toss away thermal cooling capacity by making the updated enclosure thinner ( e.g., the iMac 24" ). It wasn't so much that CPUs they had were "bad" as much as Apple designed the thermals of the enclosure to be "just enough" big enough for the nominal state of the CPU. Not designing for upper edge cases or giving themselves some 'slack' to deal with the "unexpected".

The 24" iMac is not a Pro or prosumer machine by any means. It does not have enough CPU or GPU power enough for professional computing or even prosumer computing.

If the HDMI is being added I do not think at all the enclosure of the 2021 MBP will get thinner.

It's clear now that Apple is making distinction between their consumer macs and prosumer macs.
We saw this with the 2019 Mac Pro and 2019 16" MBP. I would reserve judgement on thermal capacity of the 2021 MBP until they are properly tested by third parties.

And M2 still won't natively boot Windows (what most people have ) . Or be IT repair friendly . or .....

Most people don't care about Windows. We live in a computing age where most of work is done on the web. Apart from gaming macOS can do almost everything for the average consumer.

For the gamer of course Windows is a must. But people don't buy ultrabooks for gaming.

Most people do care about battery life, speed and support.

The avg Joe don't care about repairability.

 

[Kung gu]

It doesn't make sense for TSMC to stiff any of the high end customers they have. Apple isn't that special ( that is why Apple is paying lots of money upfront. Paying upfront doesn't make them special. Just means they paid early. Anyone else with a large enough stack of money probably can get most of what Apple is getting. )

The iPhone makes up 20% of TSMC's revenue.

The 2023 Intel grabbing a bigger chunk for volume production for ArrowLake in 2023 lines up with the ArrowLake timeline.

The core problem issue is that neither article reflect the whole lifecycle of the N3 process very well.

If there are 4-5 "big" customers ramping up for major volume products on N3, then Apple will we walking a thinner road on the ramp. It is probably not too thin but mistakes by Apple will likely not be amenable to just 'buying" their way out of . They are more likely going to cost time ( i.e., time to market ).

We don't know unless they hit the market. Intel wanted 7nm in 2016-2017. Here we are in 2021 Intel is still on 10nm.

Apple will probably do incrementally better but probably wont "take over" the general PC laptop market. Price will be an issue and frankly they won't be able order that many chips if they even wanted to.

Apple "take over" it never will. Apple needs do sell <$400 laptops do do that. BUT Apple has a huge influence in the tech industry. Other windows OEMs will take not of Apple's SoC approach and will implement it too. It's just a matter of time before Lenovo, Dell, HP start making ARM SoC laptops.

But that is a dual edge sword that doesn't necessarily go all Apple's way. Windows 11 is going to have a diverse ecosystem of suppliers. macOS won't.

That's always been the case. Look at iOS and android. The laptop space won't be any different. Apple will have the advantage of offering unique systems that no other OEM can match(eg. M1 and macOS).

 

[quarkysg]

2. You're presuming that Apple doesn't toss away thermal cooling capacity by making the updated enclosure thinner ( e.g., the iMac 24" ). It wasn't so much that CPUs they had were "bad" as much as Apple designed the thermals of the enclosure to be "just enough" big enough for the nominal state of the CPU. Not designing for upper edge cases or giving themselves some 'slack' to deal with the "unexpected".

Applying that "minimally large enough" thermals to a M1X will likely lead to similar "bad" thermals just at different levels.

Well, the story was that Intel 'promised' certain max thermals for the CPUs but failed to deliver, and Apple having originally designed their notebook enclosures with the 'promised' thermals in mind. The same could happen to Apple Silicons in theory, but I see that as unlikely, for the next few years as least.

The way I see it, designing a notebook has to be balanced between portability and power. Lugging a brick around due to thermals needs of the CPU and GPUs (more mass needed for cooling and battery) would results in poor sales of the notebook, especially for the 16" MBP. Not to mention the need to always plug it in.

 

[leman]

It doesn't make sense for TSMC to stiff any of the high end customers they have. Apple isn't that special ( that is why Apple is paying lots of money upfront. Paying upfront doesn't make them special. Just means they paid early. Anyone else with a large enough stack of money probably can get most of what Apple is getting. )

Apple is not just a rich customer to TSMC, they are also close partners who develop technology together.

 

[quarkysg]

Apple is not just a rich customer to TSMC, they are also close partners who develop technology together.

Most likely it is Apple who is pushing TSMC to advance at the speed that they did. Too bad Samsung did not foresee this or it would be their semi-conductor arm that is leading instead of TSMC.

 

[leman]

Most likely it is Apple who is pushing TSMC to advance at the speed that they did. Too bad Samsung did not foresee this or it would be their semi-conductor arm that is leading instead of TSMC.

Well, it goes beyond that - Apple actively participates in R&D. For examples, there are reports that 2nm node is being jointly developed by TSMC and Apple.

 

[UBS28]

Most likely it is Apple who is pushing TSMC to advance at the speed that they did. Too bad Samsung did not foresee this or it would be their semi-conductor arm that is leading instead of TSMC.

Samsung is also on 5nm, just like TSMC. And Apple moved away from Samsung because they are Apple’s biggest competitor.

 

[Kung gu]

Samsung is also on 5nm, just like TSMC. And Apple moved away from Samsung because they are Apple’s biggest competitor.

Samsung's 5nm is not up to parity as 5nm from tsmc

 

[leman]

Samsung is also on 5nm, just like TSMC. And Apple moved away from Samsung because they are Apple’s biggest competitor.

Apple doesn’t use Samsungs 5nm because it’s inferior even to TSMC‘s 7nm. They are more than happy to buy displays and other things from Samsung and Samsung is happy to sell because they make more money selling cutting edge tech to Apple than using it in their own phones