Discussion Intel current and future Lakes & Rapids thread

[witeken]

Now that Coffee Lake has launched, Icelake is the next stop! But when and what? With the rumored 8C Coffee Lake coming next year, I suppose will be ICL-U/Y.

 

[jpiniero]

Now that Coffee Lake has launched, Icelake is the next stop! But when and what? With the rumored 8C Coffee Lake coming next year, I suppose will be ICL-U/Y.

I was thinking, if 10 nm yields continue to be as bad as Digitimes suggests, would Intel consider cutting the EUs down to 20 for Icelake? That would allow to go to quad core while still staying pretty close to the roughly 70 mm2 die size that Cannonlake's 2+2 is. They could sell the Y parts as triple core while keeping the U at quad.

Probably would be too late in the cycle to do this.

 

[Dayman1225]

I expect Icelake to fall into 2019 if i'm honest, I hope I am pleasantly surprised, though I doubt it with Intel's recent track record and with BK at the helm. Tick Tock, Tick To-- the clock broke.

 

[witeken]

I expect Icelake to fall into 2019 if i'm honest, I hope I am pleasantly surprised, though I doubt it with Intel's recent track record and with BK at the helm. Tick Tock, Tick To-- the clock broke.

We always knew Coffee Lake would be Intel's 2018 product, but they pulled some of it into 2017. Icelake will be a 2019 product with some first SKUs (probably -U/Y) in 2018.

 

[Dayman1225]

The pulled it into 2017 and suffered stock issues, things always come at a price.

 

[jpiniero]

http://www.digitimes.com/news/a20171006PD208.html

Digitimes (gotta love 'em) has a new article, mostly on Coffee Lake but it does mention 10 nm. Between that and what Intel has said, Intel probably won't actually ship any sellable Cannonlake-Y until the very end of June. So anything 10 nm Core bigger than that 2+2 die seems unlikely in 2018.

 

[Arachnotronic]

http://www.digitimes.com/news/a20171006PD208.html

Digitimes (gotta love 'em) has a new article, mostly on Coffee Lake but it does mention 10 nm. Between that and what Intel has said, Intel probably won't actually ship any sellable Cannonlake-Y until the very end of June. So anything 10 nm Core bigger than that 2+2 die seems unlikely in 2018.

Looks like it. 14nm is shaping up to be Intel's longest-lived node: 2015 through 2018 for the bulk of its shipments.

Good thing they added Kaby and Coffee to the mix, or they'd have been in massive trouble.

 

[Ajay]

Geez, 10nm is such a train wreck. Intel is still unchallenged in laptops - so they are okay there. Desktops could be a different story. Unless the retail CFL shortage isn’t replicated at the OEM level. AMD couldn’t have timed Ryzen's release better if they tried.

 

[Ajay]

Looks like it. 14nm is shaping up to be Intel's longest-lived node: 2015 through 2018 for the bulk of its shipments.

Good thing they added Kaby and Coffee to the mix, or they'd have been in massive trouble.

I really wonder if Intel is really planning a 10nm+++ node, or if 7nm is looking real good at this point. Like you said, it’s really fortunate that 14nm turned out to have such long legs, with big help from the process development team.

 

[Arachnotronic]

I really wonder if Intel is really planning a 10nm+++ node, or if 7nm is looking real good at this point. Like you said, it’s really fortunate that 14nm turned out to have such long legs, with big help from the process development team.

I seriously doubt that if they're doing so badly on 10nm that 7nm will be any better. There was some optimism back when the 14nm node wasn't doing so well that 10nm would be better and that they'd learn from the mistakes, but now it's turning out just as badly if not worse.

 

[Dayman1225]

Maybe EUV will be the saving grace.... Bah what am I talking about, 2020 for Intels 7nm really means 2024 for Intel

 

[witeken]

http://www.digitimes.com/news/a20171006PD208.html

Digitimes (gotta love 'em) has a new article, mostly on Coffee Lake but it does mention 10 nm. Between that and what Intel has said, Intel probably won't actually ship any sellable Cannonlake-Y until the very end of June. So anything 10 nm Core bigger than that 2+2 die seems unlikely in 2018.

"Commenting on the news, Intel said that it will be shipping its first 10nm products near the end of the year beginning with a lower volume SKU followed by a volume ramp in the first half of 2018."

Intel says early 2018 and has been consistent with that timeline for quite some time now..

Edit: One thing to keep in mind, though, is that Intel is in full swing now to transition to being data center first on new nodes starting 10nm++. So if Intel is just planning on 3 generations of 10nm products, they would be in no hurry to release 10nm given 7nm will come a bit later than it would otherwise have. So for PC 10nm will have to last at least until 2022, which would be another 3-4 years or so.

Icelake full year 2019.
Tigerlake full year 2020.
?? another 10nm++(+?) product for 2021.
7nm in 2022.

 

[mikk]

Lower volume SKU means without GPU, so it won't come for notebooks. I don't count that. Mid 2018 was their plan for CNL 2+2 but this is doubtful now because digitimes claimed towards the end of 2018. It's not really important anymore though because the dualcore market will be dead by next year apart from the very low end. A new notebook with a dualcore CPU next year, no thanks. I would better buy KBL-R or wait for Icelake.

 

[jpiniero]

Mid 2018 for real 2+2 seems plausible if yields improve, but that's when they would start shipping to OEMs and not actual laptop availability. That would be later of course.

Dual core is fine; this would be the Y line mainly (eg: 5.2 W)

 

[eddman]

This is sort of like what happened with 14nm and broadwell. First broadwells were released in Sep(?)/Oct 2014 as Y parts. U parts were in Jan 2015. A few H parts in June and that's about it, I think. They gave up on broadwell.

It took intel a year to get 14nm under control, with a broad release of SKL in Sep 2015.

This pattern might apply here too. If intel manages to get CNL-Y parts out by Nov/Dec this year, we might see ICL by the end of 2018, BUT now there is 10nm+, so I don't know how it would play out. The waters are muddied.

I suppose they are not even trying to improve 10nm anymore (beyond what is needed for Y and U parts, and perhaps to make a foundation for 10nm+), and are heavily focused on 10nm+.

 

[jpiniero]

This is sort of like what happened with 14nm and broadwell. First broadwells were released in Sep(?)/Oct 2014 as Y parts. U parts were in Jan 2015. A few H parts in June and that's about it, I think.

Basically Intel won't get to Sept 2014 type yield until the end of June. That's what is being implied.

 

[mikk]

Intel's next-generation 10nm Cannon Lake processors are expected to make their debut in June or July 2018 with Cannon Lake-Y series to be released initially. Intel's 10nm Ice Lake will then be unveiled in 2019.

http://www.digitimes.com/news/a20171006PD208.html


Mid 2018 as expected, for CNL-Y at least. Ok for Y-Series Dualcore SKUs makes more sense. As for CNL-U looks like it won't come before end of 2018 (if it still comes). They should cancel this product. Digitimes didn't mention anything about CNL-U.

 

[Arachnotronic]

http://www.digitimes.com/news/a20171006PD208.html


Mid 2018 as expected, for CNL-Y at least. Ok for Y-Series Dualcore SKUs makes more sense. As for CNL-U looks like it won't come before end of 2018 (if it still comes). They should cancel this product. Digitimes didn't mention anything about CNL-U.

What a mess.

 

[Dayman1225]

So I very highly doubt we'll see a product by EOY. But we should see ramp throughout H1, hopefully. Intel, ya dun <redacted> up.

How bad is it that you cannot fab a dual core with reasonable yields with an estimated die size of 70mm^2~ They fab like 698mm^2 dies already on 14nm, albeit a very mature process, but 698mm^2 nonetheless. Come on Intel come clean on 10nm like you did with 14nm

Give us something like this, give us progress, you're only playing your self at this point Intel...

Profanity is not allowed in the technical forums
Markfw
Anandtech Moderator

 

[witeken]

So I very highly doubt we'll see a product by EOY. But we should see ramp throughout H1, hopefully. Intel, ya dun <redacted> up.

How bad is it that you cannot fab a dual core with reasonable yields with an estimated die size of 70mm^2~ They fab like 698mm^2 dies already on 14nm, albeit a very mature process, but 698mm^2 nonetheless. Come on Intel come clean on 10nm like you did with 14nm

Give us something like this, give us progress, you're only playing your self at this point Intel...

Profanity is not allowed in the technical forums
Markfw
Anandtech Moderator

Who says they have yield issues

They just launched their biggest improvement to their line-up in a decade on the same process, why would they need 10nm yet if they can milk 14nm for another year?

 

[WhoBeDaPlaya]

Who says they have yield issues

They just launched their biggest improvement to their line-up in a decade on the same process, why would they need 10nm yet if they can milk 14nm for another year?

Pffft... we're still regularly milking 180nm at work.
Hell, I taped out a couple of 0.4um designs this year.

Modern EDA tools have a surprisingly hard time dealing with (effectively due to blockages, etc.) 1.5 - 3 layer routing, eg. in Encounter, I sometimes have to actually do the routing with Wroute instead of the current NanoRoute.

 

[Dayman1225]

Who says they have yield issues

They just launched their biggest improvement to their line-up in a decade on the same process, why would they need 10nm yet if they can milk 14nm for another year?

I never said specifically yield issues, I just want them to come clean and say what is wrong with the process, not the usual "it's on track/coming" it's clear there are issues that they have yet to fix. Which isn't exactly good when you still flaunt your technical prowess with the so called 3 year lead, when they don't have a 3 year lead...

 

[witeken]

I never said specifically yield issues, I just want them to come clean and say what is wrong with the process, not the usual "it's on track/coming" it's clear there are issues that they have yet to fix. Which isn't exactly good when you still flaunt your technical prowess with the so called 3 year lead, when they don't have a 3 year lead...

They will probably do a Manufacturing Day again next year, in which they will obviously have to update on their stuff and what they have been up to. If Cannonlake isn't in the market by that time, even analysts will start asking question about where the [censored] is 10nm?

 

[Dayman1225]

They will probably do a Manufacturing Day again next year, in which they will obviously have to update on their stuff and what they have been up to. If Cannonlake isn't in the market by that time, even analysts will start asking question about where the [censored] is 10nm?

Better to question sooner rather than later... Intel have to Q2 2018 to deliver Cannonlake Y and U according to their own road map, while relatively old hopefully still holds true, any later than that and Intel is playing some sort of sick joke.

 

[DrMrLordX]

That is . . . an interesting perspective. If they CAN'T deliver the goods on 10nm, then how much longer can they pimp 14nm++ or 14nm+++ and still remain relevant? We now strongly suspect that 14nm+++ 8c Coffeelake is coming to the desktop, and I guess to the laptop world if it has to. Where does that leave them elsewhere?

Is Sapphire Rapid going to be 14nm+++? I've been told 10nm++ but I don't even see 10nm+ or 10nm yet. What about Icelake? Tigerlake?