Intel Investor Meeting 2014

[witeken]

Q&A (Diane, Kirk):

*Custom DGC SKUs have good margin.
*Very early spring launch.
*OEMs like peripherals, driving Core M.

I'm getting tired, so I might have missed some interesting things. What really bugs me is the lack of Broxton so far. I haven't even heard the name! But as BK said, they're getting more tight with the information they give (at least for process nodes; 10nm). Broxton will launch at least mid-year, so they're probably going more in-depth about that at MWC. Eul's presentation is also yet to come.

 

[witeken]

Were there any specific numbers mentioned, or a graph presented?

They had a graph with yields last year.

They revisited quite a few of the things they said at the previous IM. For the part about yields, they took the slide from 2013 and extended it further.

Was that in general, or only for certain segments (e.g. Broadwell Y Core M)? I suppose it will not be true for e.g. Broadwell-K?

It was the same slide as IDF: BDW-Y, so other 14nm products might be closer to the ~1.6x (which comes from node alone instead of including other changes with Core M).

Did they forget about Samsung 14 nm?

Also, what does it mean that competitors "pause at 16 nm"? Doesn't e.g. TSMC have 10 nm and beyond in their pipeline/plans?

Yeah, it seems like they haven't picked a FinFET Plus node. Not sure why, it might be because of practical reasons. The reason that they say pause is because if you ignore the 16/14nm node and compare 20nm with 10nm, it's the same density improvement as 1 regular node, although TSMC's and Samsung's 10nm of course could be different from the IBM data they used.

 

[witeken]

Very interesting. Where exactly are these getting posted, anyhow?

http://intelstudios.edgesuite.net/im/2014/live_im.html

 

[witeken]

So they are still giving the chips away currently. We do not know the real price OEMs are paying for atom chips but the tray price is $17 to $41. Thing is the OEMs do not pay the tray price, they pay much less than the tray price. For example according to the tray price baytrail Celerons (atom architecture) tray price goes up to $107 and the baytrail pentiums tray price go up to $161

I think OEMs pay both BOM and chip, but Intel pays the BOM overhead they faced by deciding to move quicker to market.

Very interesting by saying that Intel is admitting they are not sure they can keep their fabs at capacity with the status quo of the last 2 years. Or they are trying to deny their competitors foundry customers to reduce their profits and thus decrease their ability to do R&D to future nodes.

Sounds like they are either licensing out their IP (rockchip), or let other people use their fabs without using their IP (altera)

I think they do these partnerships just because this is a better strategy to gain smartphone market share.

--
I actually really like Intel's new density slide, which is now offset in time instead of by node, a more sensible thing to do. The slide has a lot improved in quality in those 2 months, although I'd like to hear their explanation for why TSMC's 20nm is, contrary to the slide, more dense than their 14nm Core M.

I have enjoyed the presentations so far, even though they kept quiet about some subjects. Like I predicted, it's quite a bit an update on last IM, although Kirk, at least on the process side of things.

If you haven't seen the Haiku:

Cloud forms, data shines
Transistor density sprouts
Client growth blossoms

Stacy's shroud says that the number of Haikus double every Investor Meeting. So the second one:

Big Drones, cool robots
Intelligence on the edge
BK Spawns Skynet

 

[witeken]

I was digging through my earlier post on 14nm to see what we can expect for 10nm ramp, and following is a very interesting comment indeed, from Q3'12 on 14nm:

I am not seeing anything that would cause it to be off of historical patterns. So if you look at what’s happened in odd number of years when we started the new process technology, you’ll see a increase in startup costs that hits pretty significantly in Q1, goes up some more in Q2 and then it will start to come down from there. So I am not seeing anything that would cause that to be different from historical patterns.

Guess what? I heard exactly the same here! 10nm start-up would have an impact in Q1 and Q2 and would come down in H2! Intel hasn't said anything else about 10nm. Maybe their yields are a disaster, maybe they'll wait a full 2 years after Broadwell launches this early spring to get sufficient ROI, maybe they'll make up for this year's HSW Refresh by skipping what would otherwise have been Skylake 1 year after BDW (so only 1 year of 14nm to make up for 3 of 22nm) or maybe they go in between those 2 extremes...

But the plain fact that we got confirmation that 10nm is going to be in line with historical patterns should be mind boggling.

 

[liahos1]

i thought the fact that they said 7nm would fall on the cost curve w/o euv was really interesting.

so post this analyst day do we take intel's smartphone strategy to be that they will not develop their own SOC's anymore for merchant sale and instead work only with rockchip/spreadtrum?

 

[Fjodor2001]

They had a graph with yields last year.

They revisited quite a few of the things they said at the previous IM. For the part about yields, they took the slide from 2013 and extended it further.

Hmm... so that slide is from 2013? It indicates yields on 14 nm should be same as on 22 nm in 2014Q1? Seems that did not pan out since they now are saying "Yields will match in ~H2.", assuming very late H2 since they say 14 nm yields is not comparable to 22 nm now and we're in November 2014.

I'm surprised Intel made such a bad assessment at the end of 2013 since 2014Q1 was not far away. Something must have gone really wrong with 14 nm in 2013Q4, or they painted a much brighter picture than what they knew matched reality at that point.

 

[Roland00Address]

I think OEMs pay both BOM and chip, but Intel pays the BOM overhead they faced by deciding to move quicker to market.

I was not being specific with an exact cost of Zero when I said Intel is giving away the chip. It may be negative, it may be roughly even, it may be a small profit.

Regardless charging $17 dollars for a chip, and then losing $8 to $15 in a build of materials means you are only making $2 dollars to $9 dollars per chip. The thing is we already know from sources and past experience that intel is not selling its cheapest skus at $17 dollars the real cost after bulk ordering is even less. And while atom is not paying for the R&D that creates new architectures, designs fabs, and pays those 107,600 employees (and remember many of these employees are expensive engineers) it would be nice if the department turns a profit.

But whatever as long as intel is making profits on their server divisions and their desktop line, mobile is a hobby market that may cost money but also it could turn into the next big thing.

So like I said earlier "So they are still giving the chips away currently" means the contra revenue BoM is still a significant factor. I was hoping the new intel baytrail skus would reduce the BoM to something much more affordable, but I guess we have to wait for SoFia where intel will change to a new strategy. Old Fabs and TSMC printing out good intel budget chips to capture market share, while the new shiny 14nm and someday 10nm printing out atom chips Intel makes money on. You use the old fabs to make sure you can satisfy demand while you use the new fabs to generate profit.


On another note I just realized intel 22nm atom die size is 102mm2, intel 22nm 2 core 6 eu ivybridge (pentiums and celeron type chips) is 94mm2. It actually costs intel more money to make a baytrail atom for desktops vs a ivybridge celeron or pentium for desktops. The advantage is baytrail has lower tdp and is cheaper to implement with a lot less complicated motherboard, power supply, cooling, etc. Thus Intel can actually can convince OEMs to pay a little more for the baytrail than its performance would suggest for these OEMs care about cost and not the performance with these low end skus.

 

[witeken]

Now Renée James. She gave some information about the other things that have not been talked about yet. Now introducing Hermann Eul:

*Big market (smartphones, LTE, low-end, phablets).
*Focused approach and execution: footprint in tablets, communications, foothold in entry segment.
*40M, LTE competitive, SoFIA on track.
*Share -- Value & entry: most Android; mid and premium: 50/50 with Windows.
*Showcase procucts: Dell Venue 8 thin tablet, tablet for children, Lenovo Yoga Pro 2.
*Fast TTM with reference platform for OEMs (30 ODMs). 45% of Intel tablets come from here; video shown with comments from 1 such ODM.
*Android: "it just works". Has a team to solve problems; most are bad programming, apparently.
*Moorefield Nexus player design win.
*Reference design for Android.
*2015: SoFIA ready by EOY. Master reference designs for products 3.5" to 7"+.
*Have all IP to integrate (Bleutooth, GPS, WiFi,..).
*Roadmap: BT, Cherry Trail (end of year?), Moorefield, LTE.
2016: Broxton, SoFIA 14nm. Meh.

Rockchip starts with: SoFIA 3G, Spreadtrum with SoFIA LTE (or maybe 14nm).

I'm not sure if I have heard that correctly, or if CT will release only be released EOY 2015... Edit: CT will go into production in Q1. Note that it was originally expected to launch in Q3 or Q4.

 

[liahos1]

Now Renée James. She gave some information about the other things that have not been talked about yet. Now introducing Hermann Eul:

*Big market (smartphones, LTE, low-end, phablets).
*Focused approach and execution: footprint in tablets, communications, foothold in entry segment.
*40M, LTE competitive, SoFIA on track.
*Share -- Value & entry: most Android; mid and premium: 50/50 with Windows.
*Showcase procucts: Dell Venue 8 thin tablet, tablet for children, Lenovo Yoga Pro 2.
*Fast TTM with reference platform for OEMs (30 ODMs). 45% of Intel tablets come from here; video shown with comments from 1 such ODM.
*Android: "it just works". Has a team to solve problems; most are bad programming, apparently.
*Moorefield Nexus player design win.
*Reference design for Android.
*2015: SoFIA ready by EOY. Master reference designs for products 3.5" to 7"+.
*Have all IP to integrate (Bleutooth, GPS, WiFi,..).
*Roadmap: BT, Cherry Trail (end of year?), Moorefield, LTE.
2016: Broxton, SoFIA 14nm. Meh.

Rockchip: SoFIA 3G, Spreadtrum: SoFIA LTE.

I'm not sure if I have heard that correctly, or if CT will release only be released EOY 2015... Edit: CT will go into production in Q1. Note that it was originally expected to launch in Q3 or Q4.

i dont get this I thought BK said cherrytrail production in q4. its going to take a whole year to get those to market? unbelievable.

broxton 2016 is depressing but I guess focus on the low end/value segment makes more sense in terms of being able to pick off share and not annoy apple in the process.

 

[krumme]

Good work witeken.
Perhaps Intel gives less information because they get some real competition now?
Plain and simple you dont tell excactly what you are going to do and when. But also to be able to change plans and adapt. Imo its nice to see them like that. It means they are alert, and focused-adaptive on the same time. Intel is best when they are paranoid

 

[witeken]

Hmm... so that slide is from 2013? It indicates yields on 14 nm should be same as on 22 nm in 2014Q1? Seems that did not pan out since they now are saying "Yields will match in ~H2.", assuming very late H2 since they say 14 nm yields is not comparable to 22 nm now and we're in November 2014.

I'm surprised Intel made such a bad assessment at the end of 2013 since 2014Q1 was not far away. Something must have gone really wrong with 14 nm in 2013Q4, or they painted a much brighter picture than what they knew matched reality at that point.

Who knows? Double patterning seems to be a real pain. It's mind blowing if Intel can really stay ahead of historical cost/transistor trend up to 7nm even if EUV did not exist.

Costs of 14 are still going to be pretty high until Q3. Also note that 22nm is Intel's highest yielding node ever, so matching it is a pretty good achievement I'd say.

 

[witeken]

On another note I just realized intel 22nm atom die size is 102mm2, intel 22nm 2 core 6 eu ivybridge (pentiums and celeron type chips) is 94mm2. It actually costs intel more money to make a baytrail atom for desktops vs a ivybridge celeron or pentium for desktops. The advantage is baytrail has lower tdp and is cheaper to implement with a lot less complicated motherboard, power supply, cooling, etc. Thus Intel can actually can convince OEMs to pay a little more for the baytrail than its performance would suggest for these OEMs care about cost and not the performance with these low end skus.

IB doesn't include a PCH, though.

 

[Phynaz]

Nice job, thanks for posting this.

 

[Roland00Address]

IB doesn't include a PCH, though.

Yes I agree, but the PCH is made on an older process 32nm and is roughly 80mm2 on 32nm. These labs are depreciated so if you are not using them you are wasting money and fab potential. So we are talking about 40mm2 if it was shrunk down, or be done on 32nm which needs to be used or else it is just a wasted sunk cost.

My point is what really matters on intel's profitability is how many chips they can sell, and how they can make the chips attractive to the OEMs and the higher the attraction the greater the ASP.

It is not about cutting costs on intel's end, it is instead how you can make your chips, sexy and desirable so you get what you want out of the relationship and not what the other guy (the oem and the end consumer) wants out of the relationship. Focusing on only costs, to borrow an english idiom is to be "Penny Wise and a Pound/Dollar Foolish"

 

[Idontcare]

Hmm... so that slide is from 2013? It indicates yields on 14 nm should be same as on 22 nm in 2014Q1? Seems that did not pan out since they now are saying "Yields will match in ~H2.", assuming very late H2 since they say 14 nm yields is not comparable to 22 nm now and we're in November 2014.

I'm surprised Intel made such a bad assessment at the end of 2013 since 2014Q1 was not far away. Something must have gone really wrong with 14 nm in 2013Q4, or they painted a much brighter picture than what they knew matched reality at that point.

You need to look in the file to have the correct context of that 2013 yield status slide. It was in the presentation because it was being compared and contrasted with the 2014 update to that same set of data:

What the data show is that there was indeed a marked change to the slope of the yield improvement trend starting at the end of Q3 2013. The new trend is consistent, but with lower slope throughout H1 2014.

 

[dealcorn]

Page 40 of Stacy Smith's presentation reports a forecast 2015 improvement in Mobile and Communications Group operating margin of ~800M. That improvement sounds light in relation to the current loss run rate of ~$1B/qtr. Is my comparison apples and oranges?

 

[meloz]

Does anyone believe in this? Cherry Trail this year is nonsense imho. Otherwise we would have much more infos. There is nothing. Braswell has been delayed into Q3 2015. For Cherry Trail Q2 2015 is a realistic target.

What they mean is that Intel might be ready for volume production and supply in the next 40 days. How quickly OEM/ODM can push it into market it is a different question. I don't see Q2 2015 as realistic. I expect Cherry Trail tablets to launch no later than Feburary 2015.

*Target: 60% gross margin.

Not sure whether to admire or hate.

Diane Bryant:

*Network/telco: fast move from proprietary to IA. 7.5% share.

Intel will dominate the merchant silicon market. I expect very strong growth in this segment for them, they could easily take >90% of the market if they are willing to tolerate lower profit margins.

 

[III-V]

So here's a fun fact that I think no one has caught onto:
Intel's had more design wins for Knight's Landing than their current Xeon Phi sales to date. They're already contracted for more sales of their next generation, than they've sold of their current one. Crazy.

Yeah, it seems like they haven't picked a FinFET Plus node. Not sure why, it might be because of practical reasons.

16FF+ is the same density as 16FF; it's just 15% faster. I know you participated in this discussion previously... you must have forgotten, or missed the conclusion.

Did they forget about Samsung 14 nm?

What are you referring to, exactly? That Samsung 14nm is supposedly a little denser than TSMC's 20nm? I don't think Samsung's published their 14nm parameters... and if so, that'd be why.

Page 40 of Stacy Smith's presentation reports a forecast 2015 improvement in Mobile and Communications Group operating margin of ~800M. That improvement sounds light in relation to the current loss run rate of ~$1B/qtr. Is my comparison apples and oranges?

I'm fairly certain it's apples-to-apples. Intel is going to be a LOT more competitive in mobile next year. Cherry Trail removes contra revenue, SoFIA LTE and 3G will bring tremendous improvement to their low end offerings, Bay Trail is becoming less expensive as it is (it'll cost 30% less in 2015, compared to its initial cost, I presume), and their modems are picking up adoption nicely.

i dont get this I thought BK said cherrytrail production in q4. its going to take a whole year to get those to market? unbelievable.

It's EOY '14.

 

[Nothingness]

So we have:

• Cherry Trail out by the end of the year
• Question about tablet. Moore's Law has helped. 2015: 14nm
• Roadmap: BT, Cherry Trail (end of year?), Moorefield, LTE

I guess this basically means chip prod by eoy, devices next year?

Also this claim:

Bay Trail: Chrome share from 24% in Q2'13 to 65% and lower price points. Example: Samsung's internal Exynos -> IA.

needs some balancing: ASUS and Lenovo are going to release Chromebooks with RK3288 (Rockchip with 4xCortex-A17), which is a switch from Intel to ARM, but that was an Intel presentation

PS - Witeken, thanks a lot for all the info!

 

[NTMBK]

*Bay Trail now over 50% of Celeron and Pentium.
*Bay Trail platform cost will be reduced by another 30% in 2015; allows high margins.

So Bay Trail is expected to power budget laptops and desktops, while Cherry Trail goes into tablets? Makes sense- 14nm volume is still ramping up, so prioritise the most power sensitive markets. This is the equivalent of keeping Sandy Bridge Celerons in the market while Ivy Bridge Core i(n) were already in the higher margin markets.

Did anyone ask about their use of Imagination GPUs in the future? Intel seem to be trying to improve reuse of their IP (Bay Trail forming basis of KL cores, GenX graphics in both Core and Atom, obviously Haswell cores being used from tablets up to 140W servers), and replacing licensed GPUs in mobile parts with their own tech seems like the next step.

Thanks for the write-up! :thumbsup:

 

[Idontcare]

Nice job, thanks for posting this.

+1

(and I don't do that very often )

 

[witeken]

so post this analyst day do we take intel's smartphone strategy to be that they will not develop their own SOC's anymore for merchant sale and instead work only with rockchip/spreadtrum?

Both, of course.

 

[steve wilson]

So does this info make us any wiser on when Skylake k will be released?

 

[witeken]

Not sure whether to admire or hate.

Because this were all short statements from longer explanations, I think I should clarify this one. For the last years, Intel has had a gross margin of up to 65% (although some segments like DGC are probably considerably above that). But because Intel also has their big captical spending and depreciation, this could negatively impact their gross margins, and that of course worries those investors. So in response to a question, he said that it wasn't a big issue, and they're targeting at least 60% (general, not specific for any period) gross margin.