Discussion Intel current and future Lakes & Rapids thread

[moinmoin]

I for one am sad that this means we won't see a 10nm SMESF (super mega enhanced super fin) anymore.

This was rumoured before, not surprised that Intel finally aligns their node naming with industry and it’s only 10nm ESF now being called 7, the rest aren’t.

Is the industry moving to Ångström at 2nm? Kind of oddly inconsequential to remove the "nm" for 7, 4 and 3 but add it back for 2nm, I mean 20A.

 

[Det0x]

Intel is claiming a clear path to process performance leadership by 2025, but i don't see how that can be realistic


I mean they are behind both TSMC and Samsung in the race to GAA, and that is best case, if they can stick to their own roadmap this time.

They are 2 years behind TSMC in regards to die to die stacking (hybrid bond in zen3 vcashe)

Density numbers:

IBM 2nm = Intel 20A -> (speculation)

TSMC 3nm = 292
Intel 20A = 333
TSMC "2nm" = ??? (released before intel 20A)
Screenshots taken from: Intel's Process Roadmap to 2025: with 4nm, 3nm, 20A and 18A?!

 

[JasonLD]

Is the industry moving to Ångström at 2nm? Kind of oddly inconsequential to remove the "nm" for 7, 4 and 3 but add it back for 2nm, I mean 20A.

Better marketing term than going decimals I think.

 

[moinmoin]

Intel in claiming a clear path to process performance leadership by 2025, but i don't see how that can be realistic

It was only yesterday that Intel finally stopped claiming leadership so it's actually pretty generous of them to admit that they won't have it back until at least 2025.

 

[maddie]

It was only yesterday that Intel finally stopped claiming leadership so it's actually pretty generous of them to admit that they won't have it back until at least 2025.

One sure way to get Gov attention. Give us the money and we'll put us back on top.

Also, they are going to compete with TSMC (Intel’s Foundry Services) while using TSMC production capabilities for their leading edge? Strange thing to happen.

 

[JasonLD]

Also, they are going to compete with TSMC (Intel’s Foundry Services) while using TSMC production capabilities for their leading edge? Strange thing to happen.

TSMC 3nm rumor itself could be a mixed up rumor now that Intel 7nm has become 3/4nm.

 

[maddie]

Can someone explain this line from Ian's article? I would have expected differently.

"EUV will only intercept Intel’s portfolio with its new Intel 4 process, where it will be used extensively, mostly on the BEOL."

 

[Hitman928]

Can someone explain this line from Ian's article? I would have expected differently.

"EUV will only intercept Intel’s portfolio with its new Intel 4 process, where it will be used extensively, mostly on the BEOL."

Back End Of Line. Basically the metal layers.

Front End of Line is for poly/wells/etc. This is the general break down between the two.

 

[2blzd]

Ian saying Alder Lake is "now on 7nm" is confusing. It's still 10nmESF, just relabled as Intel 7...but throughout the article he kept saying ADLS is 7nm.

 

[maddie]

Back End Of Line. Basically the metal layers.

Front End of Line is for poly/wells/etc. This is the general break down between the two.

Yeah I know that, but EUV mainly used for the BEOL? Shouldn't it be used mainly for the smaller constructs in the FEOL?

 

[Asterox]

Intel is claiming a clear path to process performance leadership by 2025, but i don't see how that can be realistic
View attachment 47802

I mean they are behind both TSMC and Samsung in the race to GAA, and that is best case, if they can stick to their own roadmap this time.
View attachment 47803
They are 2 years behind TSMC in regards to die to die stacking (hybrid bond in zen3 vcashe)

Density numbers:
View attachment 47804
IBM 2nm = Intel 20A -> (speculation)

TSMC 3nm = 292
Intel 20A = 333
TSMC "2nm" = ??? (released before intel 20A)
Screenshots taken from: Intel's Process Roadmap to 2025: with 4nm, 3nm, 20A and 18A?!

It kind of reminds me of .....................

 

[RanFodar]

Well, does anyone expect Intel to rise back, especially with new leadership? Not that I believe in their "unquestioned leadership", but its possible to reach that level of parity with their competitors. Good for the consumer, eh?

 

[gdansk]

I am cautiously optimistic. If only because Intel still failing by 2025 could be disastrous for consumers (I mean me). It seems they have an agreement with ASML (and IBM?) to help them achieve it.

They could always compete with AMD as a fabless company too. It probably isn't vital to Intel that they stay a manufacturer. Alder Lake looks like it may be competitive while including more features on a similar process to AMD's Zen 3. Albeit a bit late but that was process problems.

 

[eek2121]

You're kidding. Please don't ever work in a position where you give advice to users on what to buy in order to browse just fine.

I didn't bother reading other comments, since the Intel announcement has blown up the thread. Instead, I came here to say this and I'm too busy enjoying summer to wade through them all.

Boy do I have news for you.

I happen to work on a major healthcare web application that, assuming you live in the US, probably has your information buried in our database(s). If you live in 47 of the 50 current U.S. states or one of the territories and have EVER approached the pharmacist for anything this is guaranteed. Prior to that, I've been involved with very large successful eCommerce sites (thank goodness not involving Digital River, and mostly not Amazon). We also have a contract that covers many EU citizens. The eCommerce stuff I worked on is in use worldwide.

I also have done a ton of hands on time in the IT world. If you've ever worked with a data center in NJ or a handful of other states, you might be dealing with me work.

I've done a ton of hardware engineering and even some biology stuff as well.

Finally, I have my hands dirty in the open source world as well.

All of that is just scratching the surface. I've been in hardware and software engineering since the 90s.

If you use the internet, you have probably use something of mine. I'm not trying to troll you (well mostly, but when you call into question my professionalism with that comment...), I call into question yours. What have YOU done for the world? I didn't invent the internet, shoot, I didn't invent most of the stuff I worked on, but my name is attached to more than a few projects that are well known today.

(The funny part is I don't even do it for money. I'm not rich...I just enjoy engineering...I'm probably foolish at this point for not making a harder play on a startup...but I just want to build cool stuff).

Yes... the internet is usable on a raspberry pi. Dollars to donuts says I could give you access to the internet on both an RPi 4 and a budget PC, both running Linux and the same DE and you wouldn't be able to tell the two apart without checking. For bonus points, I could give you access to a Pi 4 and a medium to high end laptop and you ALSO wouldn't be able to tell the difference.

Outside of that, just glad Intel is finally getting smart. They killed the process ridicule (mostly, with a rebrand), now they have to show up with the products. That second part will be the hard part. I personally don't think Alder Lake will do it for the desktop. I think Alder Lake is their next "Core" laptop moment, and the successor is a "Core 2" desktop moment. I also think Alder Lake will only bring them to parity with Zen 3. They will be behind Zen 4 for a while.

If they mess up next-gen, you know marketing is still running the company along with boneheaded execs.

 

[eek2121]

Ian saying Alder Lake is "now on 7nm" is confusing. It's still 10nmESF, just relabled as Intel 7...but throughout the article he kept saying ADLS is 7nm.

Well, that is because Intel 10nm is not TSMC “10nm”. Please note that I am not supporting Intel with this comment, but rather, reminding folks once again that TSMC and Intel are not all that far apart when it comes to process. TSMC is on “5nm” currently, and current released Intel stuff is on the equivalent of TSMC “7nm”. Intel 7nm is probably around TSMC “4nm”. Intel 7 is around TSMC “5nm”, but likely behind “N5P” and 7+ is likely smaller.

I wish we had standardized silicon benchmarks for all of them.

Also, Nostra, before you post, don’t. I love you, but we’ve been there, done that, and I walked away with the t-shirt. 🤣😊😉

 

[Doug S]

Well, that is because Intel 10nm is not TSMC “10nm”. Please note that I am not supporting Intel with this comment, but rather, reminding folks once again that TSMC and Intel are not all that far apart when it comes to process. TSMC is on “5nm” currently, and current released Intel stuff is on the equivalent of TSMC “7nm”. Intel 7nm is probably around TSMC “4nm”. Intel 7 is around TSMC “5nm”, but likely behind “N5P” and 7+ is likely smaller.

I wish we had standardized silicon benchmarks for all of them.

What difference would standardized benchmarks make? Neither Intel's old names or the new names had any basis in physical reality, not for a long time.

When the node names did conform to physical reality everyone tried to scale by sqrt(2) every two years as that would double the number of transistors which fit Moore's Law and what ITRS dictated in its technology roadmap for semi equipment. They used various strategies to reduce leakage, increase drive current, etc. etc. that would allow higher clock rates but the process names were based solely on transistor density.

So if you want a "standardized benchmark" compare them by transistor density. Unless there is someone in this forum who actually makes decisions ordering wafers on a leading edge process, the node names only useful for Intel vs AMD vs Apple fanboying.

 

[Gideon]

If intel can execute on this roadmap they will be quite competutive. However a relevant tweet how performance numbers don't quite add up for retaking the crown just yet:

https://twitter.com/x/status/1419777135779135496

 

[uzzi38]

Well, that is because Intel 10nm is not TSMC “10nm”. Please note that I am not supporting Intel with this comment, but rather, reminding folks once again that TSMC and Intel are not all that far apart when it comes to process. TSMC is on “5nm” currently, and current released Intel stuff is on the equivalent of TSMC “7nm”. Intel 7nm is probably around TSMC “4nm”. Intel 7 is around TSMC “5nm”, but likely behind “N5P” and 7+ is likely smaller.

I wish we had standardized silicon benchmarks for all of them.

Also, Nostra, before you post, don’t. I love you, but we’ve been there, done that, and I walked away with the t-shirt.

> Intel 7 is probably behind N4

> Intel 7 is probably around N5 but behind N5P

Uh, which one is it?

Anyway, I don't really mind the naming changes all that much. Although having to refer to 10ESF as 7nm now is just plain weird. And I don't expect Intel's 4nm to be measurably more dense than N5/N5P products, given how Intel tend to spam HP cells more than everyone else, with product densities tending to be smaller than the competition's (e.g. Lakefield compute tile is 49MTr/mm^2, despite being a die with no I/O)

 

[mikk]

Based on this slide, 2 Gracemont clusters (8 cores) is the size of 2x Golden Cove Cores or 1/4 the size

View attachment 47793

If this is accurate roughly, we can say Gracemont is still the small core (as expected from most people). So it's not like they had to increase Gracemont to Skylake size level.

If intel can execute on this roadmap they will be quite competutive. However a relevant tweet how performance numbers don't quite add up for retaking the crown just yet:

https://twitter.com/x/status/1419777135779135496

They are claiming to regain the lead with High-NA-EUV which is coming with Intel 18A (presumably in 2025) and not Intel 3 which isn't even RibbonFET and PowerVia based. There are no numbers for Intel 20A, probably too early to tell. High-NA-EUV can be a game changer, Intel is the first customer they said.

 

[coercitiv]

Yes... the internet is usable on a raspberry pi. Dollars to donuts says I could give you access to the internet on both an RPi 4 and a budget PC, both running Linux and the same DE and you wouldn't be able to tell the two apart without checking. For bonus points, I could give you access to a Pi 4 and a medium to high end laptop and you ALSO wouldn't be able to tell the difference.

I can understand how the way @lobz replied may have triggered a bit of engineering pride in you, but to equate "usable" to "cannot differentiate in a blind test" is a really poor way to conduct an argument.

 

[JasonLD]

If intel can execute on this roadmap they will be quite competutive. However a relevant tweet how performance numbers don't quite add up for retaking the crown just yet:

https://twitter.com/x/status/1419777135779135496

I don't think that is how it really works.



From N7 to N5, you either get 15% performance increase at the same power or reduce power 30% at same transistor speed.
From N5 to N3, you get 10~15% performance at the same power or reduce power by 25~30%
You can't just inverse the number of certain percentage of less power to performance/watt increase.

 

[mikk]

Intel also revealed the GPU EU count for Meteor Lake yesterday:

Up to 192 EUs


That's slightly over 2x the EU count of current TGL-U. According to earlier rumors it's based on Gen12_7 which is Gen12 HP(G) based instead of Gen12 LP, but this is not confirmed. If it's Gen12HP uarch there might be IPC and feature improvements. Being based on Gen12_7 could make sense because Gen12 LP only goes up to 92 EUs, with Gen12HP(G) they can simply reuse a lower end version of it which scales from 96-512 EUs from what we know.

 

[lobz]

I didn't bother reading other comments, since the Intel announcement has blown up the thread. Instead, I came here to say this and I'm too busy enjoying summer to wade through them all.

Boy do I have news for you.

I happen to work on a major healthcare web application that, assuming you live in the US, probably has your information buried in our database(s). If you live in 47 of the 50 current U.S. states or one of the territories and have EVER approached the pharmacist for anything this is guaranteed. Prior to that, I've been involved with very large successful eCommerce sites (thank goodness not involving Digital River, and mostly not Amazon). We also have a contract that covers many EU citizens. The eCommerce stuff I worked on is in use worldwide.

I also have done a ton of hands on time in the IT world. If you've ever worked with a data center in NJ or a handful of other states, you might be dealing with me work.

I've done a ton of hardware engineering and even some biology stuff as well.

Finally, I have my hands dirty in the open source world as well.

All of that is just scratching the surface. I've been in hardware and software engineering since the 90s.

If you use the internet, you have probably use something of mine. I'm not trying to troll you (well mostly, but when you call into question my professionalism with that comment...), I call into question yours. What have YOU done for the world? I didn't invent the internet, shoot, I didn't invent most of the stuff I worked on, but my name is attached to more than a few projects that are well known today.

(The funny part is I don't even do it for money. I'm not rich...I just enjoy engineering...I'm probably foolish at this point for not making a harder play on a startup...but I just want to build cool stuff).

Yes... the internet is usable on a raspberry pi. Dollars to donuts says I could give you access to the internet on both an RPi 4 and a budget PC, both running Linux and the same DE and you wouldn't be able to tell the two apart without checking. For bonus points, I could give you access to a Pi 4 and a medium to high end laptop and you ALSO wouldn't be able to tell the difference.

Outside of that, just glad Intel is finally getting smart. They killed the process ridicule (mostly, with a rebrand), now they have to show up with the products. That second part will be the hard part. I personally don't think Alder Lake will do it for the desktop. I think Alder Lake is their next "Core" laptop moment, and the successor is a "Core 2" desktop moment. I also think Alder Lake will only bring them to parity with Zen 3. They will be behind Zen 4 for a while.

If they mess up next-gen, you know marketing is still running the company along with boneheaded execs.

So you wrote this book out of misdirected anger and unwarranted insecurities about your professionalism? I really don't care if you work IT or botanics, if you say something that makes zero sense. It has nothing to do with your professional standards, how did you even arrive at that topic? The way people (other than grandmas who were coerced into it by their families) generally want to browse the internet is not the way you set up your blind tests or your strawberry Pi experiments. Browsing multiple modern websites (while not pausing everything else in your life and on your device) for an extended period of time (eg. you don't have an OCD and you just have opened things laying around sometimes) seamlessly can be a difficult task for most last-gen budget PCs and laptops, let alone a blueberry Pi.
Since there's no benchmark for that, only the 25+ years of suffering in agony whenever I had to do anything on most of other people's machines, your unquestionable professionalism provides nothing to your point.
You might be the case of a field-leading expert with not much sense for real world operations.

 

[lobz]

Hmm, maybe, though i think having a comparable performance is more important than theoretical density maximum anyways.

Not from a business standpoint though.

 

[lobz]

Ian seems to be saying that the 20A is the IBM node albeit doesn't directly say that.

Is Intel going hip and names its node like a true redditor? Intel 20 AYYYBM

(To the trigger heavy people: please understand that this is a joke and I know they didn't name that process in the first place)