Why is Intel still using 14nm ?

[Gerard]

Intel 14nm is better than everyone elses 14nm

Because everyone else has moved on, and as said before, intel have had plenty of practice at 14nm given they've been stuck on it since around 2015.

Because they got lazy, complacent and didn’t think AMD would ever be a threat again, boy were they wrong

They were ALMOST right as AMD were on the verge of going out of business a few times in that lean period, but they put their eggs in the one basket and have been scrambling (badum tish) ever since.

 

[DarrenM343]

Because they got lazy, complacent and didn’t think AMD would ever be a threat again, boy were they wrong

I don't disagree with that but I think Intel are dong so well still on 14nm their competitors should be afraid of maybe what Intel can do on 7nm or 5. Maybe there's no relationship but I suspect Intel throwing loads of R&D at it, so when they are finally able to do 7nm or 5, it's gonna be good.

Really like what AMD are dong but would be foolish to write off Intel IMO, something quite a few seem to be doing now.

Intel did seem to be caught with their pants down and quite rightly been dealt a severe kicking

 

[the wopster]

I don't disagree with that but I think Intel are dong so well still on 14nm their competitors should be afraid of maybe what Intel can do on 7nm or 5. Maybe there's no relationship but I suspect Intel throwing loads of R&D at it, so when they are finally able to do 7nm or 5, it's gonna be good.

Really like what AMD are dong but would be foolish to write off Intel IMO, something quite a few seem to be doing now.

Intel did seem to be caught with their pants down and quite rightly been dealt a severe kicking

Intel have tried to to go to 10nm a few times now and have failed every time

 

[Mesai]

Intel 14nm is better than everyone elses 14nm

At this point they're basically just competing against themselves

 

[Cyber-Mav]

They done well. If they went to 7nm overnight i doubt anyone could compete with them.

 

[4K8KW10]

Don't forget these "nm" labels are somewhat arbitrary measurements and actually worthless for comparing between different companies.

This is wrong and false.

The nm scale indicates the smallest feature size on the transistor.
All nm are subject to compliance to the industry standard.

This is why TSMC doesn't have a 7nm node, it has a marketing N7 node.

 

[Hotwired]

This is wrong and false.

The nm scale indicates the smallest feature size on the transistor.
All nm are subject to compliance to the industry standard.

This is why TSMC doesn't have a 7nm node, it has a marketing N7 node.

Exactly how are you disagreeing with me?

 

[Mesai]

They done well. If they went to 7nm overnight i doubt anyone could compete with them.

They can't though so they're partially starting from scratch.

 

[sandys]

At this point they're basically just competing against themselves

The process shouldn't matter to the end consumer, price and performance should be the only considerations, 7 or 5nm doesn't make something better automagically

At 14nm Intel is still competitive on core performance with 7nm parts, why wouldn't you continue to use it, they have the the fabs, it can produce the chips on those fabs, whilst refining its smaller nodes, seems like a decent enough reason to still use it, I'm sure a number of the fabless companies are wishing they could produce their own chips right now.

 

[Mesai]

The process shouldn't matter to the end consumer, price and performance should be the only considerations, 7 or 5nm doesn't make something better automagically

At 14nm Intel is still competitive on core performance with 7nm parts, why wouldn't you continue to use it, they have the the fabs, it can produce the chips on those fabs, whilst refining its smaller nodes, seems like a decent enough reason to still use it, I'm sure a number of the fabless companies are wishing they could produce their own chips right now.

Because they are at their absolute limit and you can only do so much before you need shrink so that you can scale up. They have to change their process and architecture, and I don't see how they're going to manage that while staying competitive, at least not initially.

They might still perform OK, but everything else is miles worse than the competition, which is partially why they lost Apple and why they're going to keep slipping in the server and mobile market.

 

[Gerard]

Don't forget these "nm" labels are somewhat arbitrary measurements and actually worthless for comparing between different companies.

That's always bugged me, you'd think there would be a standard "measurement" that would have to be adhered to. But right now its just company x claiming they're on 7nm with company y saying their 10nm is comparable to company x's 7nm etc.

 

[philo-sofa]

I mean... Intel is on 14nm simply because it can't get process nodes below that to work at volume yet. It doesn't want to, it has to because 14nm its only option for volume desktop CPUs right now.

 

[Gerard]

I mean... Intel is on 14nm because it can't get process nodes below that to work at volume yet. They don't want to, they just have to because it's all they have for volume desktop CPUs.

It's that simple.

Makes me wonder what they're doing wrong, to be stuck on 14nm for 5 years is a little odd. A cpu is much less complex than a gpu as well so doesn't say much for their fabs if tsmc\samsung can produce something with billions more transistors with relative ease.

 

[philo-sofa]

Makes me wonder what they're doing wrong, to be stuck on 14nm for 5 years is a little odd. A cpu is much less complex than a gpu as well so doesn't say much for their fabs if tsmc\samsung can produce something with billions more transistors with relative ease.

Yeah it's a far cry from the Intel we knew of old. They always used to lead on process node shrinks.

If I had to guess it'd be a combo of the same lethargy we've seen in terms of engineering chip designs (driven by failboat, non-engineering bosses) coupled with the genuine difficulty of engineering such small processes. But given Samsung, TSMC and presumably others have managed it fine, I'd say it's mainly poor management.

 

[sandys]

Makes me wonder what they're doing wrong, to be stuck on 14nm for 5 years is a little odd. A cpu is much less complex than a gpu as well so doesn't say much for their fabs if tsmc\samsung can produce something with billions more transistors with relative ease.

I'm not sure you could call a CPU less complex than a GPU, a GPU contains lower speed more architecturally simple cores that are targeted towards specific tasks and then they stuff loads of them on the die, you can see that with the die shots, basically copy and paste of a ton of units, something that is probably simpler to put on an MPW and test.

When you make memories for example you can tune the process for the structure you are developing and can rely on high level of replication providing you with data consistency which is why when you layout an SRAM etc, you can pretty much break all the design rules as you have been around the loop a few time and qualified your cell through test vehicles etc, you can also add in a high level of redundancy, GPU manufacturers employ similar tactics due to number of small units.

 

[KompuKare]

Yeah it's a far cry from the Intel we knew of old. They always used to lead on process node shrinks.

If I had to guess it'd be a combo of the same lethargy we've seen in terms of engineering chip designs (driven by failboat, non-engineering bosses) coupled with the genuine difficulty of engineering such small processes. But given Samsung, TSMC and presumably others have managed it fine, I'd say it's mainly poor management.

Or you could look at like this:
What happens if a company thinks the high-volume lower-margin mobilephone market it beneath them?
And continuously crippled Atom so it wouldn't compete with Core and Xeon?
Well, one thing which seemed to have happened is a few years after the smartphone boom and with a possible tablet boom, Intel panicked so much they wrapped every Atom tablet with $20+ in the infamous contra-revenue as they tried to dump their into the market.
In the meantime, TSMC had larger and larger volumes and re-invested the proceeds into more and more R&D and fabs.
Back in the day when x86 was truly the junk ISA, Intel's volumes enabled them to outcompete first their CISC rivals (imagine Motorola's 68000 had 32 bit registers in the early 1980s while Intel had a segmented mess with 64KB pages etc), and then later their RISC rivals.
Volumes are extremely important.
So a lot the rot at Intel set in when they forget that turning down volume business is a very risky thing.

 

[d_brennen]

I wouldn't mind the industry staying on 7nm for another year/18months while Apple runs away with all the 5nm supply. Let's Intel catch up 7nm TSMC must be well trodden now as well

 

[philo-sofa]

Or you could look at like this:
What happens if a company thinks the high-volume lower-margin mobilephone market it beneath them?
And continuously crippled Atom so it wouldn't compete with Core and Xeon?
Well, one thing which seemed to have happened is a few years after the smartphone boom and with a possible tablet boom, Intel panicked so much they wrapped every Atom tablet with $20+ in the infamous contra-revenue as they tried to dump their into the market.
In the meantime, TSMC had larger and larger volumes and re-invested the proceeds into more and more R&D and fabs.
Back in the day when x86 was truly the junk ISA, Intel's volumes enabled them to outcompete first their CISC rivals (imagine Motorola's 68000 had 32 bit registers in the early 1980s while Intel had a segmented mess with 64KB pages etc), and then later their RISC rivals.
Volumes are extremely important.
So a lot the rot at Intel set in when they forget that turning down volume business is a very risky thing.

It's interesting, I was listening to Jim Keller on the Lex Fridman podcast, and he said what they did wrong on the mobile front was having a monolithic 'build it thus and they shalt come' approach which was only valid if you had the best possible design out there. Whereas ARM gave general guidelines in their chips but allowed their IP partners to deliver significantly different chips based off their designs. They then got to learn what works best as well as allowing chips to be built for specific tasks in the moment: both contributed to efficiency. He said Intel's Atom and mobile centric chips were actually quite good, but they never got true take-up because of that 'one design' philosophy and lacked the ability to respond to such a changing market due to that same philosophy.

That having been said, I don't know that Intel lacked revenue or commitment to mobile, nor did they lack revenue for process node improvement. They just plain messed up shrinking beyond 14nm (and even getting to that was difficult).

 

[Moeks]

Well they have had a shake up now, new CEO and what not, make no mistake, Intel have the funds to hire whoever they think they need to get this sorted if they are still having problems. We need both Intel and AMD to exist in relative balance (one is always likely to have some sort of an advantage at something at any given point in time), neither one running too far away from the other, its how the CPU market will stay healthy and interesting.

 

[turbot1984]

The future's bright, the future's 7nm lol
https://bit-tech.net/news/tech/cpus/meteor-lake-will-be-intels-first-7nm-cpu/1/