Discussion Intel current and future Lakes & Rapids thread

[jpiniero]

Honestly cannot see that happening, especially since its more than likely CNL-Y releasing first, it'll be useless without a IGPU.

Maybe there would be an industrial use for it. I think it was more so BK could say that he did get something on 10 nm released in 2017. Either way, yeah, low volume.

 

[Dayman1225]

Yes, and it's not really "22nm FFL" at all. It shares more in common with their 14nm than it does with their 22nm! Excellent for their custom foundry. - David Schor

Patience. (If you're wondering about yield, they did not talk about anything related to yield curve/learning. I even asked that in person and was told they cannot discuss it at this time.) - David Schor

So looks like we're waiting for Investors Day and MFG Day in 2018 for that yield curve/learning info....

 

[Yotsugi]

So looks like we're waiting for Investors Day and MFG Day in 2018 for that yield curve/learning info....

Sound suspicious.

 

[Dayman1225]

Sound suspicious.

Yah don't say!

 

[Dayman1225]

I'm not really sure where to put this, and I don't want to create a new thread just for this -

BARCLAYS GLOBAL TMT CONFERENCE
Dan McNamara, Corp. VP & GM of Intel's Programmable Solutions Group, Intel Corporation

Around 7:50 he begins to talk about their 10nm products

"Working on 10nm for 2 years now, we're going to sample to customers in 18"

"Very good engagement model for 10nm with the customers, good perf/watt, in time for major growth sectors"

"Altera is a sequential company, 20nm delays led to 14nm delays, but when acquired by Intel they "Power walled" (I think that's what he said, wasn't very clear) to 10nm."

"Aimed at cloud, DC and efficiency, going at it with a complete portfolio and FPGA is apart of that"

"I think GPUs were adopted for AI/DC because they were already there, FPGAs are infinitely programmable. We have accelerated our software stack and are close to seamless integration"

Seems to be Falcon Mesa, early adoption of HBM3 and EMIB 2nd gen.

 

[Dayman1225]

Erm not exactly. 10 and 10+ are being developed concurrently (I confirmed it at IEDM). I wouldn't necessarily extrapolate the 14nm trend onto the 10nm roadmap which is already confusing enough because Intel wants to shift data center a full cadence forward. - David Schor

As a side note. I am no longer convinced yield is actually that bad after some of the rumors I've heard... - David Schor

2016 was a delay, yes. But I'm slowly reaching the conclusion that 2017 appears to be a strange case of "14nm is much more profitable so we can wait a bit longer". Not a yield problem 😉 - David Schor

 

[jpiniero]

^^

If that were the case, Intel would have released Cannonlake-Y by now (and not June launch as the rumors are suggesting) since they don't have a comparable Coffee Lake product available.

Whiskey Lake I'm assuming would just be U.

 

[Dayman1225]

^^

If that were the case, Intel would have released Cannonlake-Y by now (and not June launch as the rumors are suggesting) since they don't have a comparable Coffee Lake product available.

Whiskey Lake I'm assuming would just be U.

It's rumor vs rumor now it seems. Though CNL - Y wouldn't offer anything massive over KBL - Y, so it's fine if they didnt actually refresh it with CFL, WHL etc. As he mentioned Intel seems to be holding back because Intel's 14nm is more profitable obviously. KBL - Y is already a low volume 14nm product, the only reason it has volume is due to Apple, really. IMO they should just scrap CNL all together....

 

[Yotsugi]

IMO they should just scrap CNL all together....

I don't think BK will do that, ever.
It's basically equal to Intel admitting that they fucked up.

 

[raghu78]

I don't think BK will do that, ever.
It's basically equal to Intel admitting that they fucked up.

Yeah Intel has already delayed CNL to the point that when it arrives in H2 2018 nobody is going to be bothered about it. CFL 4C+3e will embarass CNL in 15w notebooks. CNL will get a symbolic token launch in Q4 2018 before ICL starts ramping in H1 2019.

 

[DrMrLordX]

They're delaying 10nm because it's less profitable? WTF kind of thinking is that? So many of their product lines are stalled out, backed up waiting for node advancements. Phi is basically dead, Cannonlake is dead, they won't go past 6c on the desktop and 28c on the server as they continue to pimp 14nm+/14nm++. The notebook sector is a confusing mess. I can't clearly identify any upcoming Intel notebook/laptop products that amount to an improvement over what they have on the market today.

Not too long ago, they pushed billions out the door to try to advance into a new sector (mobile). Now they can't be arsed to use a "less profitable" node to keep their existing businesses afloat and at least vaguely on schedule?

 

[Dayman1225]

They're delaying 10nm because it's less profitable? WTF kind of thinking is that? So many of their product lines are stalled out, backed up waiting for node advancements. Phi is basically dead, Cannonlake is dead, they won't go past 6c on the desktop and 28c on the server as they continue to pimp 14nm+/14nm++. The notebook sector is a confusing mess. I can't clearly identify any upcoming Intel notebook/laptop products that amount to an improvement over what they have on the market today.

Not too long ago, they pushed billions out the door to try to advance into a new sector (mobile). Now they can't be arsed to use a "less profitable" node to keep their existing businesses afloat and at least vaguely on schedule?

They have a road map to follow of course, the majority if not all 10nm products are in 2019 (assuming CNL is axed), there is no point in releasing a non competitive product that will not sell simply coz "we wuz 10nm now". CNL Y/U will get killed by KBL R, CFL U and WHL U... Instead of releasing a product they could show us a few wafers, and chips that work + a yield curve/learning graph like they did with their 14nm

 

[Yotsugi]

They're delaying 10nm because it's less profitable? WTF kind of thinking is that? So many of their product lines are stalled out, backed up waiting for node advancements. Phi is basically dead, Cannonlake is dead, they won't go past 6c on the desktop and 28c on the server as they continue to pimp 14nm+/14nm++. The notebook sector is a confusing mess. I can't clearly identify any upcoming Intel notebook/laptop products that amount to an improvement over what they have on the market today.

Not too long ago, they pushed billions out the door to try to advance into a new sector (mobile). Now they can't be arsed to use a "less profitable" node to keep their existing businesses afloat and at least vaguely on schedule?

It's a bad excuse and you know it.

They have a road map to follow of course, the majority if not all 10nm products are in 2019 (assuming CNL is axed), there is no point in releasing a non competitive product that will not sell simply coz "we wuz 10nm now". CNL Y/U will get killed by KBL R, CFL U and WHL U...

BK promised us 10nm.
Now he must deliver (or eat hit pants live on T&M Day).

 

[DrMrLordX]

They have a road map to follow of course, the majority if not all 10nm products are in 2019 (assuming CNL is axed), there is no point in releasing a non competitive product that will not sell simply coz "we wuz 10nm now". CNL Y/U will get killed by KBL R, CFL U and WHL U... Instead of releasing a product they could show us a few wafers, and chips that work + a yield curve/learning graph like they did with their 14nm

They had roadmaps showing multiple 10nm products in 2017/2018 but they didn't follow those.

Regardless, Intel is still a technology company, no matter what other ideas are floating around inside BK's head. They have to continue updating products to stay relevant to the market. OEMs/system integrators are only going to put up with respins/nospins for so long. Big-boy buyers shopping through ODMs had Skylake server products months before us plebs got to see it . . . what is Intel offering them now?

Google: So what have you got for us this time? Skylake is getting long in the tooth.
BK: How 'bout Skylake on 14nm++? This time with more Optane!
Google: So about those discounts . . .

The whole point to releasing 10nm now is to remain relevant to the market. 14nm++ is nice, but the market is moving forward with or without Intel. That 10nm CNL Y/U is going to get "killed" by 14nm+/++ products is Intel's fault. It's even worse that I still can't see how CLU U or WHL U really improve Intel's mobile offerings. At least KBL mobile products improved battery life and/or max turbo some over SKL mobile products.

It's a bad excuse and you know it.

It actually looks worse than "10nm is messed up so we have to wait to 2019 to use it". Maybe not to investors who can be duped, but anyone with a long-term interest in Intel's market position can see that voluntarily avoiding 10nm on a cost basis is suicidal.

In any case, you know it's bad when the excuse is worse than the truth.

 

[Yotsugi]

The whole point to releasing 10nm now is to remain relevant to the market.

Y SKUs are pretty irrelevant, and CNL-S, CNL-EP/EX are canned, CNL-U is probably also awaiting the axe.
Like, they can't offer anything worth buying on 10 to anyone.
Heck, they need to pray to the Dark Gods for GloFo to somehow fuck up the execution on 7LP, so their 10+ products will be earlier than the competition.

 

[raghu78]

They're delaying 10nm because it's less profitable? WTF kind of thinking is that? So many of their product lines are stalled out, backed up waiting for node advancements. Phi is basically dead, Cannonlake is dead, they won't go past 6c on the desktop and 28c on the server as they continue to pimp 14nm+/14nm++. The notebook sector is a confusing mess. I can't clearly identify any upcoming Intel notebook/laptop products that amount to an improvement over what they have on the market today.

Not too long ago, they pushed billions out the door to try to advance into a new sector (mobile). Now they can't be arsed to use a "less profitable" node to keep their existing businesses afloat and at least vaguely on schedule?

The reasoning that 10nm is delayed because its less profitable than 14nm is absolute BS being pushed by Intel supporters who cannot explain the seemingly never ending 10nm delays from 2016 to 2017 to H2 2017 to H2 2018 (and god knows if there are more delays in store). By that same reasoning Intel should have kept on delaying 14nm because 22nm is their highest yielding node ever. The 8th generation notebook lineup is a chaotic mess of different generations including 14+ KBL-R, 14++ CFL and 10 CNL, the last of which is probably going to never be made in significant volume. Intel's product pipeline has been thrown into chaos because of the multiple delays to 10nm.

 

[ehume]

Y'know, an unusable 10nm is going to be less profitable than usable 14nm, 14nm+ or 14nm++ no matter how little profit they make.

 

[Kaloi48]

According to lastest information from PTT forum, ICL-U/Y 4+2 is scheduled to release between February-May 2019.

Prod: ww51'18-12'19
RTS: ww07-20'19

 

[jpiniero]

According to lastest information from PTT forum, ICL-U/Y 4+2 is scheduled to release between February-May 2019.

That does match up with Whiskey Lake roughly a year earlier. Does make it look like that H/S would be a 2H 19 product.

 

[DrMrLordX]

I thought Whiskey Lake was just going to be Coffee Lake on a different platform?

 

[IntelUser2000]

According to lastest information from PTT forum, ICL-U/Y 4+2 is scheduled to release between February-May 2019.

So they want to launch Whiskey Lake close to BTS season and launch Icelake 6 months later? What does one offer the other? If they see the launch can be close because it uses the same PCH, it still doesn't make sense because laptops are refreshed as a whole, not on a chip by chip basis. Manufacturers are going for one and ignore the other, and Intel will be forced to do the same.

Looking at the translated page, they also claim there are other variants of KBL-G(or whatever it is called now) with GDDR5 instead of HBM.

 

[Dayman1225]

So they want to launch Whiskey Lake close to BTS season and launch Icelake 6 months later? What does one offer the other? If they see the launch can be close because it uses the same PCH, it still doesn't make sense because laptops are refreshed as a whole, not on a chip by chip basis. Manufacturers are going for one and ignore the other, and Intel will be forced to do the same.

Looking at the translated page, they also claim there are other variants of KBL-G(or whatever it is called now) with GDDR5 instead of HBM.

WHL is apparently just a 4+2 version of CFL U (4+3e) so I guess WHL U would be the more widely adopted version, with Icelake Y/U refreshing both Y and U parts with Gen11 GPU, 10nm+ and the new ICL arch.

 

[IntelUser2000]

WHL is apparently just a 4+2 version of CFL U (4+3e) so I guess WHL U would be the more widely adopted version, with Icelake Y/U refreshing both Y and U parts with Gen11 GPU, 10nm+ and the new ICL arch.

It sounds to me you are saying there's WHL and WHL-U? There's only WHL-U according to that site. WHL-U is a replacement for KBL-R with the extremely limited information we have. KBL-R launched around August this year, so BTS 2018 for WHL-U makes sense.

Again according to that information, Icelake Y/U is coming 6 months later. Perhaps Icelake is only a two core part, and that's why we need Whiskey Lake. Otherwise, Whiskey Lake is a very short lived part that'll be ditched by most manufacturers. Or, as I have said there's something in Whiskey Lake that doesn't exist in Icelake.

Parts like Whiskey Lake and the rumors of 8 core Coffeelake must exist because there's no way Intel could have any Icelake parts in 2018, yet they need new product to sell.

 

[Dayman1225]

It sounds to me you are saying there's WHL and WHL-U? There's only WHL-U according to that site. WHL-U is a replacement for KBL-R with the extremely limited information we have. KBL-R launched around August this year, so BTS 2018 for WHL-U makes sense.

Again according to that information, Icelake Y/U is coming 6 months later. Perhaps Icelake is only a two core part, and that's why we need Whiskey Lake. Otherwise, Whiskey Lake is a very short lived part that'll be ditched by most manufacturers. Or, as I have said there's something in Whiskey Lake that doesn't exist in Icelake.

Parts like Whiskey Lake and the rumors of 8 core Coffeelake must exist because there's no way Intel could have any Icelake parts in 2018, yet they need new product to sell.

I meant WHL U and only that, as it's apparently just a 4+2 version of CFL U (Which is 4+3e), in the conversation they mention 4+2 Icelake, not 2+2.

 

[IntelUser2000]

I meant WHL U and only that, as it's apparently just a 4+2 version of CFL U (Which is 4+3e), in the conversation they mention 4+2 Icelake, not 2+2.

Yea, that's why I see no point of WHL-U mere 6 months before Icelake. No sane manufacturer is going to get that over Icelake.