Discussion Intel current and future Lakes & Rapids thread

[Eug]

There are now 3 client roadmaps, with the one of coercitiv above is actually named 'Desktop' while the WCCF ones are named 'Mobile' and 'Commercial'

Hmmm

 

[coercitiv]

Yes but I still don’t understand. The Desktop roadmap includes a 5W Y series timeline. What are these for?

It's not a Desktop roadmap, WCCF made a mess there: it's a B2B roadmap which includes both mobile and desktop offerings with a special life cycle meant to maximize availability and compatibility. (SIPP)

Just focus on the previously posted Mobile roadmap and the newly leaked Desktop roadmap.

 

[Eug]

Ok. For Ice Lake Y dual-core limited in 2019, who do you think the “limited” customer would be? The timing would work well for Apple as a customer, since they haven’t updated their 12” MacBooks for 2 years, and a dual-core refresh would make sense in terms of their model segmentation.

 

[IntelUser2000]

Ok. For Ice Lake Y dual-core limited in 2019, who do you think the “limited” customer would be?

It's probably just top selling models for every manufacturer, rather than one manufacturer taking it all. HP is probably one since they are the one that uses the custom on-package LTE modem chip.

The timing would work well for Apple as a customer, since they haven’t updated their 12” MacBooks for 2 years, and a dual-core refresh would make sense in terms of their model segmentation.

With all the talks of them abandoning Intel chips, they shouldn't focus too much on Apple as a customer. I don't mean to give up on them, but it would be a mistake to give to them all in expense of others.

 

[Ajay]

On Ice-Lake from 1Q19 transcript:

We also previewed our upcoming 10 nanometer Ice Lake client CPUs, which will deliver unprecedented levels of integration including DL Boost Inference acceleration, Wi-Fi 6, Thunderbolt 3 and Gen 11 graphics. Our first integrated GPU with a full teraflop of performance. Our 10 nanometer yields continue to improve and Ice Lake remains on track to be in volume systems on retail shelves for the 2019 holiday selling season In Q4...

 

[Eug]

It's probably just top selling models for every manufacturer, rather than one manufacturer taking it all. HP is probably one since they are the one that uses the custom on-package LTE modem chip.

With all the talks of them abandoning Intel chips, they shouldn't focus too much on Apple as a customer. I don't mean to give up on them, but it would be a mistake to give to them all in expense of others.

Well, the reason I asked is because it seems that Core i5-8210Y is effectively an Apple exclusive, and that only came out 2018 Q3, with retail availability mid Q4 2018.

 

[DrMrLordX]

Desktop Roadmap

No desktop 10nm parts through the end of 2020? What the hell, Intel? That is the worst-case scenario coming to life. They don't even have Rocket Lake there, it's Comet Lake. Zen3 vs rehashed Skylake on 14nm?

And those Xeons. Woof.

On Ice-Lake from 1Q19 transcript:

So Intel is indeed not going to have IceLake U/Y available until Christmas. That's what I thought. The reports of said chips being available for back-to-school 2019 were apparently false.

 

[Markfw]

No desktop 10nm parts through the end of 2020? What the hell, Intel? That is the worst-case scenario coming to life. They don't even have Rocket Lake there, it's Comet Lake. Zen3 vs rehashed Skylake on 14nm?

And those Xeons. Woof.



So Intel is indeed not going to have IceLake U/Y available until Christmas. That's what I thought. The reports of said chips being available for back-to-school 2019 were apparently false.

Yes, I don't really care to even speculate on Intel, until they have at lease ONE decent new product.

 

[jpiniero]

On Ice-Lake from 1Q19 transcript:

The first Intel TF IGP is wrong in any case. The Skylake GT4e is up to 1152 GF.

 

[JDG1980]

Unless Comet Lake and Rocket Lake are going to be backports of the Sunny Cove architecture, I don't see how Intel is going to effectively compete on the desktop once Zen 2 comes out. They can't push clocks any further than they already have, and Zen 2 is going to mostly (if not entirely) close the gap on IPC while having significantly better perf/watt and higher core counts. If these leaked roadmaps are real, Intel might even have to compete with Zen *3* with yet more 14nm+++++ Skylake rehashes. They will need a breakthrough on par with Conroe or Summit Ridge to get back into the game after this epic fail.

 

[jpiniero]

It's probably just top selling models for every manufacturer

No, more like highest priced models, well at least for the quads. Like the X1 Carbon that was mentioned in that Lenovo leak. Volume won't be high enough for anything Apple or even Surface (except for the Lakefield product)

The dual cores, we'll have to see. Probably won't see those until the Comet Lake update with LPDDR4 is available; which will make it easier on the OEMs to supply both in a model.

 

[coercitiv]

Unless Comet Lake and Rocket Lake are going to be backports of the Sunny Cove architecture, I don't see how Intel is going to effectively compete on the desktop once Zen 2 comes out.

Comet Lake is Skylake based as Tom's Hardware reports.

That changed in early March 2019 as the Comet Lake PCI CPUIDs, used for identifying the processor to an operating system, were listed on freedesktop.org, an organization that helps ensure interoperability for Linux-based operating systems. This posting gave us little information to work on, aside from the note that the processors wielded Intel's Gen9 graphics, which implies the chips come with some variant of the 14nm process. The listing also states that Comet Lake "comes off" the Coffee Lake family, which suggests the chips feature that same microarchitecture.

Rocket Lake on the other hand looks like it's built on chiplets, the different roadmaps show it sporting 14nm graphics in the commercial offering and 10nm graphics in the mobile client offering. The CPU is still on 14nm though, but given the change to chiplets and new support for PCIe Gen 4 and being based on ICL PCH... this should be Sunny Cove. (see a nice analysis here)

I remember this forum being rebuked back in Dec 2018 for "not getting the memo" on how Intel already had the tools in place for a 2019 port of Sunny Cove on 14nm. Turns out context still beats the most optimistic interpretation of an interview reply.

 

[jpiniero]

this should be Sunny Cove.

Much more likely Willow Cove if it's not Skylake.

 

[repoman27]

So there are two three roadmaps here (Client Commercial / SIPP, Client Mobile, and Desktop Client), all leaked by tweakers.net, but I have to agree with jpiniero's initial assessment. These do not appear to be genuine documents from either Intel or Dell (as the Dell logo on the mobile and desktop roadmaps would suggest). However, they could be recreations made by someone with access to the originals in an attempt to obscure the source. The biggest red flag is the red squiggly spelling checker underlines on the SIPP roadmap beneath the words Kaby and Mehlow. That would mean that what we're looking at is a screen-shot taken while those elements were still stored as editable text and the document was open in an editor where check spelling as you type was enabled.

The other problem is that while typos and errata are not unheard of in these documents, there are more than could be considered normal here. The client commercial document is a 3-year roadmap covering 2019 to 2021, yet the heading says "(2018-2021)". Amber Lake Y is listed as "2/4C" on the first two roadmaps, but is strictly a 2C part (and Intel knows this because it's literally the same die as KBL-Y 2+2). On the other hand Ice Lake Y is listed on the client mobile roadmap as "2C" when it should be 2/4C. Rocket Lake U is listed twice in the client commercial document, once on the U-Series line and again on the H / G Series line.

So maybe this is just a hasty transcription of actual roadmaps with a Dell logo pasted on one of them as a false-flag attempt. Overall, a lot of what is presented seems quite plausible. However, the left-hand edges of the roundrect boxes on roadmaps provided by Intel are generally accurate down to the working week, even if they're nowhere near where we'd like to see them. That accuracy is unfortunately not present in these documents either, as demonstrated by some of the boxes for parts that have already been released, when even older leaks were spot-on.

edit: the "152 of 14" and partially obscured Dell logo on the "Desktop Client" (not "Client Desktop") roadmap also strike me as a bit odd.

 

[Eug]

The other problem is that while typos and errata are not unheard of in these documents, there are more than could be considered normal here. The client commercial document is a 3-year roadmap covering 2019 to 2021, yet the heading says "(2018-2021)". Amber Lake Y is listed as "2/4C" on the first two roadmaps, but is strictly a 2C part (and Intel knows this because it's literally the same die as KBL-Y 2+2). On the other hand Ice Lake Y is listed on the client mobile roadmap as "2C" when it should be 2/4C. Rocket Lake U is listed twice in the client commercial document, once on the U-Series line and again on the H / G Series line.

I could see Ice Lake Y in 2019 for a "limited" product being dual-core only.

The biggest red flag is the red squiggly spelling checker underlines on the SIPP roadmap beneath the words Kaby and Mehlow. That would mean that what we're looking at is a screen-shot taken while those elements were still stored as editable text and the document was open in an editor where check spelling as you type was enabled.

? Even as a PowerPoint slide, those red squiggles would appear.

 

[Yotsugi]

I could see Ice Lake Y in 2019 for a "limited" product being dual-core only

Intel promised us -U, and that one is definitely 4c.

 

[Eug]

Intel promised us -U, and that one is definitely 4c.

It says 2/4C on the same roadmap for Ice Lake U, and it says it's in 2019.

We were specifically talking about Ice Lake Y.

 

[repoman27]

I could see Ice Lake Y in 2019 for a "limited" product being dual-core only.

Agreed, but Amber lake isn't about to grow another pair of cores, and it's pretty much a given that ICL-Y 4+2 and ICL-U 4+2 will be released sometime this quarter. ICL-Y was always rumored to be 4+2. I don't think I'm being overly optimistic here, given material statements made by Intel to investors. Intel can only ease 14nm supply pressure once they can reallocate wafer starts that had been going to WHL-U 4+2 and KBL/AML-Y 2+2.

? Even as a PowerPoint slide, those red squiggles would appear.

If that text was editable and you took a screenshot of it, yes, but that's exactly the kind of scenario I was describing. The other would be if you converted a .pdf to an editable document by opening it with Microsoft Word. These types of materials are rarely circulated as editable documents. Official PowerPoint slide decks would only contain images.

 

[jpiniero]

ICL-Y was always rumored to be 4+2

They intend there to be a 4+2 Y, but of course given how bad the functional yields must be, getting enough QC chips that also pass the quality bin needed for Y seems tough.

 

[repoman27]

Also, Comet Lake Y 2/4C in Q3'19 for consumer and Q2ish'20 for SIPP... is CML-Y really going to be a thing? Maybe ICL-Y 4C didn't work out, but have we seen any other mention of CML-Y?

 

[jpiniero]

Also, Comet Lake Y 2/4C in Q3'19 for consumer and Q2ish'20 for SIPP... is CML-Y really going to be a thing? Maybe ICL-Y 4C didn't work out, but have we seen any other mention of CML-Y?

Since they are doing Comet-U, may as well do Comet-Y.

 

[NostaSeronx]

My very accurate prediction is the best in the world, yuge and accurate believe me.

Intel is doing 7-nm in 2020, with newly acquired R&D from 2016/2017.

This will allow Intel to have 18-nm minimum metal pitches on 193i. While, all the competitors are stuck at 28-nm or 24-nm minimum metal pitches, and extremely decreased throughput with EUV.

The architecture will be heavily derived from the SoftMachines architecture. With the Global Front End((Instruction Bus)& Back End(Data Bus)) being >10 years ahead of what it was in Shasta. Utilizing IP not used till now on 7-nm, etc. SoftMachines says 2x-3x across HPC to IoT, while Intel's futuristic improvements will bump that number up to 4x-6x across Data-centric and User-centric usage.

 

[Ajay]

I was taking a second look at Intel's Q1 earnings call a ran across this exchange:

Harlan Sur -- JPMorgan -- Analyst
Good afternoon and thanks for taking my question. Just wanted to get an update on 10 nanometer manufacturability. Last quarter, the team mentioned 10 nanometer yields were tracking 14 nanometer yields at a similar point prior to production ramp. Is the team still seeing good improvements in 10 nanometer yields? Are you still tracking 14 nanometer yield ramps? And can you just give us an update on early 7 nanometer development and manufacturability?

Venkata S. Murthy Renduchintala -- Chief Engineering Officer & Group President of Technology, Systems Architecture & Client Group
Harlan, hi, this is Murthy, I'll take that question. I can only add to what Bob said in his opening statements that we continue to make solid progress against our plan that we shared with you during the course of 2018. And as I said on the last call, I feel better about our traction today than I did 90 days ago. So that continues to bode well for our product launch ambitions, which Bob summarized as having systems on shelf for holiday season in 2019 with a barrage of products across all of our businesses to follow shortly thereafter.
And I would like to take the opportunity to just remind everybody that at CES and in the analyst meeting we had the end of last year, we did show 10 nanometer across the entire portfolio of our product ranges. We talked about Ice Lake clients, which clearly was top of mind in the early discussion. So we also talked about Lakefield. Bob mentioned that as well. Navin talked about 10 nanometers for Ice Lake server. And we also talked about 10 nanometers moving into our network in the 5G program, which we believe is going to be a big growth sector.
So the story is not just about 10 nanometer yields but 10 nanometer now being a key part of our entire product portfolio. And as I say, I think that coupled with our focus on the pillars of technology that Bob talked about, in my mind I think puts our product portfolio looking forward in a pretty good position. So net-net, I think 10 nanometers is looking better now than at the last earnings call. It's broadly deployed across our portfolio. And that in combination with the other technology ingredients that Bob talked about, we believe, sets us up for a pretty exciting product roadmap.

1). So, Intel is pretty bullish about 10nm (Swan also cites 10nm ramp as a significant capital drain). I can’t see how Intel can make these claims by C-level executives (who have fiduciary responsibility) without risking the prospect of serious investor:backlash if they prove false.

2). At this point, I can only see Icelake server CPUs existing with the use of EMIB. It appears, as there is no client 'S' series CPUs being announced, that relatively small core Icelake chiplets are the only option for Intel to make competitive high core count CPUs.

What say ye, forum denizens?

 

[Dayman1225]

I was taking a second look at Intel's Q1 earnings call a ran across this exchange:


1). So, Intel is pretty bullish about 10nm (Swan also cites 10nm ramp as a significant capital drain). I can’t see how Intel can make these claims by C-level executives (who have fiduciary responsibility) without risking the prospect of serious investor:backlash if they prove false.

2). At this point, I can only see Icelake server CPUs existing with the use of EMIB. It appears, as there is no client 'S' series CPUs being announced, that relatively small core Icelake chiplets are the only option for Intel to make competitive high core count CPUs.

What say ye, forum denizens?

I say that I am cautiously optimistic, something about those 'leaked' roadmaps just doesn't feel right, especially with the typos most notable for me is listing Skylake SP as 4 channel and "Cascade Lake Refresh" which doesn't exist according to Intel's own public roadmap... who knows...

edit: Remind that Intel is holding an Investor Day on the 8th of May, we might hear a little bit more about 10nm health when that happens.

 

[jpiniero]

I say that I am cautiously optimistic, something about those 'leaked' roadmaps just doesn't feel right, especially with the typos most notable for me is listing Skylake SP as 4 channel and "Cascade Lake Refresh" which doesn't exist according to Intel's own public roadmap... who knows...

Cascade Lake Refresh is likely Cooper Lake. That's one issue with that roadmap, even if legit, it's likely out of date. One theory is that they took Cascade Lake Refresh and (because of Rome) updated it to be dual die using EMIB.

I doubt you will see Icelake Server actually make it to the wild even in a scenario where yields magically improved to say mediocre instead of terrible. Once Rome ships, any big die Xeon less than 32 cores is going to be unsellable. I at this point think the Icelake Server that's been floating around is still monolithic.