3nm Intel chips

[UBS28]

Looks like TMSC will produce 3nm chips for Intel laptops and server CPU's. And the rumours are, they have a bigger contract than Apple so Apple will only use their 3nm chips for the iPads.

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

Intel could be back sooner than most people expect.

 

[dmccloud]

It's nowhere near as neat and simple as that. Based on some remarks from Intel themselves, they will only be using 3nm for certain products in its Xeon lineup (which could include some thin and light notebooks). As for the bulk of their upcoming consumer products (Meteor Lake and Granite Rapids), those will be pushed to the 7nm process in late 2022. Intel could book enough 3nm capacity to fulfill orders for the consumer market as a whole even if Apple wasn't already moving to 3nm themselves. Chances are that Apple will be using 3nm for both the iPad and iPhone, so they will be using a LOT more of TSMCs capacity than you might think.

Tom's Hardware

 

[leman]

I doubt that Intel can outbid Apple for foundry access, so it’s probably not that simple. Also, we still have no idea how Apple chip lineup will look going forward. There is a good chance that M1 is a one-off product used to kickstart the entire process and that the future models will look very differently from their phone mobile counterparts. I can for example see Apple leveraging 3nm in phones and tablets for power efficiency, while keeping the desktop chips on 4nm for a while.

 

[Falhófnir]

That may be for their planned SiFive Risc V chips though? Seems they're planning on forging their own path that way, which if recent history with the x86 hegemony is anything to go by will probably be good in case Arm chips begin to stagnate as well (though Arm and Risc V seem to be quite similar so maybe they will run into similar problems at similar times?). I'd say the writing is on the wall for x86 at this point. Certainly for consumer computing, I guess enterprise stuff might cling on longer.

 

[Kpjoslee]

That may be for their planned SiFive Risc V chips though?

That is going to be fabricated on Intel 7nm as their pipecleaner. It wouldn't make sense to outsource the product they are contracted to manufacture themselves lol.

 

[KPOM]

It's nowhere near as neat and simple as that. Based on some remarks from Intel themselves, they will only be using 3nm for certain products in its Xeon lineup (which could include some thin and light notebooks). As for the bulk of their upcoming consumer products (Meteor Lake and Granite Rapids), those will be pushed to the 7nm process in late 2022. Intel could book enough 3nm capacity to fulfill orders for the consumer market as a whole even if Apple wasn't already moving to 3nm themselves. Chances are that Apple will be using 3nm for both the iPad and iPhone, so they will be using a LOT more of TSMCs capacity than you might think.

Tom's Hardware

Though it does say that Intel’s 3nm orders will be larger than Apple’s. TSMC is in an interesting position with Intel. Since the latter has made no secret that it is doubling down on its foundry business, TSMC is basically helping a future competitor. It seems like they wanted a big commitment from Intel. Apple needs about 200 million chips annually for its iPhone, iPad, and Mac lines. Not all need to be 3nm.

 

[T'hain Esh Kelch]

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

Is CPU architecture something where you can just dial up or down on the scale? Or will porting to such a smaller node, be a major re-architecture?

 

[MrGunny94]

Is CPU architecture something where you can just dial up or down on the scale? Or will porting to such a smaller node, be a major re-architecture?

Look at how Tiger Lake got back port.. There's some interesting info about that, however for the question in hand these 3nm changes will be accompanied by a new arch, all previous gen Intel SoCs were designed to be manufactured at Intel plant so they would have to adapt to TSMC.

 

[falainber]

That may be for their planned SiFive Risc V chips though? Seems they're planning on forging their own path that way, which if recent history with the x86 hegemony is anything to go by will probably be good in case Arm chips begin to stagnate as well (though Arm and Risc V seem to be quite similar so maybe they will run into similar problems at similar times?). I'd say the writing is on the wall for x86 at this point. Certainly for consumer computing, I guess enterprise stuff might cling on longer.

Intel had a contract with TSMC for 3nm process way before they signed SiFive agreement so it is highly unlikely that 3nm will be used for Risc V. Besides, RISC V just does not need such advanced process.

 

[Kung gu]

And the rumours are, they have a bigger contract than Apple so Apple will only use their 3nm chips for the iPads.

This is not true. Apple will use 3nm in iPhone 14 using the the A16 SoC. The 3nm SoC will also be used in iPads and Macs. Also the last 8 months Apple sold OVER 100 MILLION iPhone 12's. The iPhone is a bigger market than intel laptops.

 

[GMShadow]

Is CPU architecture something where you can just dial up or down on the scale? Or will porting to such a smaller node, be a major re-architecture?

It generally requires a re-think even across similar size nodes at different fabs. The A9 was (in)famously dual-sourced from TMSC and Samsung, which meant Apple had to make design changes to have the chip work on each process.

 

[Serban55]

Looks like TMSC will produce 3nm chips for Intel laptops and server CPU's. And the rumours are, they have a bigger contract than Apple so Apple will only use their 3nm chips for the iPads.

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

Intel could be back sooner than most people expect.

Not until 2024...Apple has around 78-80% of TMSC production until Q3 2024
So, while i think we all be glad that Intel will produce someday 3nm...the time doing that will be a big factor
Until we see the light of that day when Intel will deliver 3nm..we should not take it for granted
And i doubt that Apple will use next year 2 or 3 different dies/architecture for ipads iphones and for macs
From business and profit margins stand points Apple showed they can use the same SoC for their ipads and "entry" levels macs such as macbook air, mac mini

 

[Significant1]

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

It was/is. Apple is also on the 3nm train and designing it own chips mean it can add custom hardware that fits macOS. Anything they can offload to hardware rather than software, will be better performing and more power efficient.

 

[Bug-Creator]

so it is highly unlikely that 3nm will be used for Risc V. Besides, RISC V just does not need such advanced process.

You "need" an advanced process if you need to squeeze more transistors/performance out of a given space/energy restraint.
Wether that is x86, RISC-V or their GPUs really depends on where they want to sell these in a few years.

As for Intel (x86) outperforming AS, the M1 being 5nm over Intels 10/14nm is just a part of the advantage so all a 3nm Intel would do is reduce the gap.

Also given that Intel does try to market it's fab services and is gonna build new fab themselves suggest that this either a short term solution or only aimed at some smaller markets where it just doesn't make sense for Intel to build it's own fab.

"bigger order" means nothing without knowing the details, it may very much be that TMSC plans for Apple's bleeding edge to be on 3nm+ or 2nm by the time Intels contract gears up.

 

[Apple Knowledge Navigator]

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

The size of the fabrication process isn't the be-all indicator for how 'good' a chip is.

In fact, the smaller die sizes Apple has achieved thus far are merely a side-effect of the transition. The biggest factors were to do with control over the design of the chip, the release schedule, and the integration of hardware-software throughout the entire product eco-system.

 

[apparatchik]

Well, seeing that last year's iPhone's A14 chip is already faster in single-threaded perf than any of Intel's designs / chips ever produced while consuming a fraction of the power, I wouldn't be too optimistic as the process solving by itself Intel's situation.

Apple is beating them hard on absolute perf, but it's complete humiliation in efficiency or perf per watt, the process will help in these regards for sure but only up to a certain point, when Intel & Apple where on the same fabrication process a few years back, Intel couldn't deliver a chip performant/efficient enough for phones & tablets, remember Atom? I would be extra cautious about day dreaming Intel's future.

 

[leman]

Well, seeing that last year's iPhone's A14 chip is already faster in single-threaded perf than any of Intel's designs / chips ever produced while consuming a fraction of the power, I wouldn't be too optimistic as the process solving by itself Intel's situation.

To be fair however, current premium Intel and AMD (later only on desktop) SKUs match or slightly outperform M1 in single-threaded performance. Power efficiency is another matter entirely, true, but the bragging rights mostly come from performance.

So if Intel can port their current designs to TSMC 3nm, they should be able to easily reach 20-30% higher performance while slightly reducing the power consumption. That will be more than enough to counter M1, at least on paper. Yes, Intel would still run hotter, but that’s arguably less of an issue, especially since they will have hybrid CPU cores to get some of that battery life back.

The big question is what Apple will deliver, as I doubt they will be standing still. I’m very exited to see how Intel and Apple CPUs will compare when produced at the same node.

 

[MacModMachine]

Looks like TMSC will produce 3nm chips for Intel laptops and server CPU's. And the rumours are, they have a bigger contract than Apple so Apple will only use their 3nm chips for the iPads.

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

Intel could be back sooner than most people expect.

wont matter , apple buys fab far in advance. they already are locked in. intel wouldent have the clout to push apple out.

 

[CWallace]

Going from 14nm --> 3nm will yield huge performance and energy efficiency gains for Intel, up to the point you might have to wonder if going for their iOS based chips was a wise choice.

I am supremely confident Intel/TSMC's first 3nm chip is not going to be a Core-i7 or Core-i9 so I doubt Apple has any concerns about moving to Apple Silicon on the Mac...

 

[Kpjoslee]

This is not true. Apple will use 3nm in iPhone 14 using the the A16 SoC. The 3nm SoC will also be used in iPads and Macs. Also the last 8 months Apple sold OVER 100 MILLION iPhone 12's. The iPhone is a bigger market than intel laptops.

If TSMC 3nm is ready to start production by 2H 2022, it would be too late for A16. Reason the article states A16 would end up using 4nm. And how Intel is able to secure decent 3nm capacity because of that.

Apple probably prefers to use latest TSMC node for their A series SoC but they wont delay iPhone release just for that.

 

[pshufd]

To be fair however, current premium Intel and AMD (later only on desktop) SKUs match or slightly outperform M1 in single-threaded performance. Power efficiency is another matter entirely, true, but the bragging rights mostly come from performance.

So if Intel can port their current designs to TSMC 3nm, they should be able to easily reach 20-30% higher performance while slightly reducing the power consumption. That will be more than enough to counter M1, at least on paper. Yes, Intel would still run hotter, but that’s arguably less of an issue, especially since they will have hybrid CPU cores to get some of that battery life back.

The big question is what Apple will deliver, as I doubt they will be standing still. I’m very exited to see how Intel and Apple CPUs will compare when produced at the same node.

Apple is running at 3.2 Ghz while Intel and AMD are running at 5 Ghz. It's pretty clear that Apple cares more about power efficiency than absolute single-core performance.

 

[leman]

Apple is running at 3.2 Ghz while Intel and AMD are running at 5 Ghz. It's pretty clear that Apple cares more about power efficiency than absolute single-core performance.

They clearly care about both. After all, that 3.2 GHz CPU is as fast as Tiger Lake running at 5ghz. My point is that power efficiency alone will not be enough for the prosumer hardware. If Apple does not manage to completely outclass every single comparable all PC hardware offering, Apple Silicon will be a failure.

 

[pshufd]

They clearly care about both. After all, that 3.2 GHz CPU is as fast as Tiger Lake running at 5ghz. My point is that power efficiency alone will not be enough for the prosumer hardware. If Apple does not manage to completely outclass every single comparable all PC hardware offering, Apple Silicon will be a failure.

That's absurd. Success is determined by maximizing shareholder value and you don't need the fastest CPU in the world to do that.

 

[leman]

That's absurd. Success is determined by maximizing shareholder value and you don't need the fastest CPU in the world to do that.

There are different ways to measure success. But if the 2022 Apple Silicon MacBook Pro 16" ends up being slower than a 2022 Intel-based Dell XPS 15", that's not going to be good PR for Apple, and it certainly won't give the users a lot of confidence in Apple's ability to deliver pro-level hardware.

At any rate, I expect prosumer Apple Silicon to be 20% faster than whatever Intel or AMD will come up with, while using half the power or less. And Intel's potential utilization of TSMC 3nm won't change this.

 

[pshufd]

There are different ways to measure success. But if the 2022 Apple Silicon MacBook Pro 16" ends up being slower than a 2022 Intel-based Dell XPS 15", that's not going to be good PR for Apple, and it certainly won't give the users a lot of confidence in Apple's ability to deliver pro-level hardware.

At any rate, I expect prosumer Apple Silicon to be 20% faster than whatever Intel or AMD will come up with, while using half the power or less. And Intel's potential utilization of TSMC 3nm won't change this.

The X86 decoder penalty is insurmountable which is why everyone is looking at ARM or RISC.