Intel Investor Meeting 2015: November 19

[frozentundra123456]

So how about a translation into plain english of what Bill Holt said. Is the cost still going down with each node, how low can we go, when will we see ultraviolet lithography and next gen materials?

 

[witeken]

Thanks for the write up! I think describing themselves as not a PC company is a little premature when it makes up 60% of their revenue...

How they describe it is that it is their core business that gives them scale, but not for growth, that comes from NAND and data center and IoT.

 

[witeken]

So how about a translation into plain english of what Bill Holt said. Is the cost still going down with each node, how low can we go, when will we see ultraviolet lithography and next gen materials?

He didn't say concrete things about next-gen things and EUV.

Yes, cost per transistor will still come down by >0.69x.

But that assumes yields are on par, of course.

 

[witeken]

* Growth rate less than expected (even down revenue) because weak GDP and end of XP refresh
* Enterprise down a bit; HPC, networking and storage, workstation up 10%; cloud up over 40%
* In 2016 will be about 1/3 split between the three
* Cost increase as 14nm ramps
* Client forecast: a year ago slightly up, actually about 8% down
* Client revenue down 10%, operating profit down 27% because of 14nm
* ASP down ~5% till 2009, then flat/up because of shift in mix
* About $1B improvement in mobile profitability (contra-revenue gone and other things)
* No LTE ramp yet
* Capex down a lot: 10nm delay, efficiencies on 14nm, volume down
* 2016 capex up: 10nm + China factory
* Cost of platforms: will be up significantly in performance in 2016 because of transition to 14nm; maintstream and value will be back heading to 2014 levels in Q4'16 after very high 2015
* Gross margin: up because of ASP in data center; down because of 10nm and 3D XPoint ramp

 

[mikk]

Do you know if slides are available?

Some slides are here: http://www.computerbase.de/bildstrecke/69065/3/

 

[kimmel]

I dunno.

You should really look at the slides when they become available.

http://intelstudios.edgesuite.net/im/2015/pdf/2015_InvestorMeeting_Bill_Holt_WEB2.pdf

slide 19

30% of the die is sram on A9
9% of the die is sram on SKL 2+2

That is a pretty big difference. Also A9 doesn't have reg files.

 

[witeken]

The 14nm yield projection still seems awefully optimistic. It's been since the start, when they thought it would be on par in Q1'14 instead of Q4'13.

It just seems to asymptote.

 

[Hans de Vries]

Yeah, I can't really describe a whole slide, let alone in a few seconds.

But no. This 0.7x is the naive density comparison. The whole point of these foils from Holt is to show that you are not comparing apples to apples. You have to normalize the die composition.

If you do that, you'll see that Intel is 1.4x denser than 20nm BEOL. Just like theoretically predicted in 2013!

(Anyone remember the A8 and Core m comparison of 2B transistors vs 1.3B? It naively suggests Apple's 20nm is vastly denser, but in fact it's the 1.4x number as bad.)

The slides can be viewed in the video around 2h15m

They are shockingly ignorant really, worthless.

One example, they still manage to claim that Samsung's density at 14nm is
the same as at 20nm. When is someone going to tell them that is nonsense?


Samsung 20nm to 14nm

contacted gate pitch     90nm   -->  78nm
metal layer 1 pitch      80nm   -->  64nm
Exynos 5433 vs 7420      113mm2 -->  78mm^2

Density increased by a factor 1.45 which is nicely illustrated by
the scaling of the 20nm Exynos 5533 to the 14nm Exynos 7240.

This is common knowledge but not to the people who should know apparently.

http://www.avsusergroups.org/jtg_pdfs/JTG2015_7JamesRevSm.pdf

 

[witeken]

Data center

* 15% CAGR through 2019: HPC 20%, cloud >20%, comms >20%, enterprise ~flat.
* Most is CPU, in 2019 something like 20% will be non-CPU.
* Some of network, HPC and enterprise becomes cloud.
* TAM expansion + increased ASP
* 3/4th of cloud is software as a service
* 2/3rd is consumer, 1/3 business
* Number of big players using custom CPUs, silicon photonics, FPGA, 3D XPoint interest.
* Network: $17B TAM, 10% IA MSS (rest is ASSP, FPGA, ASIC), up from 5% MSS in 2013; growing at 10x the market.
* AT&T, Verizon, Vodaphone, Verizon, SK Telecom moving to IA.
* Photonics: on silicon, integration, lower power and form factor and cost -- production early 2016
* Omni-Path Architecture: $1.6B HPC TAM: performance (latency), power and cost. Will be MCP and then on-die.
* 3D XPoint: $34B TAM in 2020: 4x capacity of DRAM, 0.5x cost -- for Skylake platform
=> Want to be leadership in all three above.

Next up: Kirk Skaugen of client and mobile

 

[tarlinian]

The slides can be viewed in the video around 2h15m

They are shockingly ignorant really, worthless.

One example, they still manage to claim that Samsung's density at 14nm is
the same as at 20nm. When is someone going to tell them that is nonsense?


Samsung 20nm to 14nm

contacted gate pitch     90nm   -->  78nm
metal layer 1 pitch      80nm   -->  64nm
Exynos 5433 vs 7420      113mm2 -->  78mm^2

Density increased by a factor 1.45 which is nicely illustrated by
the scaling of the 20nm Exynos 5533 to the 14nm Exynos 7240.

If you accept those numbers for Samsung 20 nm, then it's not any denser than Intel's 22 nm node. (Those metal and gate pitches are exactly the same as Intel's 22 nm.) They did have a published 20 nm process with 64 nm metal pitch...I don't know if any products were released with this.

 

[Arachnotronic]

The slides can be viewed in the video around 2h15m

They are shockingly ignorant really, worthless.

One example, they still manage to claim that Samsung's density at 14nm is
the same as at 20nm. When is someone going to tell them that is nonsense?


Samsung 20nm to 14nm

contacted gate pitch     90nm   -->  78nm
metal layer 1 pitch      80nm   -->  64nm
Exynos 5433 vs 7420      113mm2 -->  78mm^2

Density increased by a factor 1.45 which is nicely illustrated by
the scaling of the 20nm Exynos 5533 to the 14nm Exynos 7240.

This is common knowledge but not to the people who should know apparently.

http://www.avsusergroups.org/jtg_pdfs/JTG2015_7JamesRevSm.pdf

Samsung 20nm LPP, which never actually made it into production, was basically equivalent to TSMC 20nm.

 

[witeken]

Thoughts on IM Arachnotronic?

 

[witeken]

Client: PC and mobile

* Large old installed base
* New form factors, user experiences and segmentation, thinner, faster, etc.
* Skylake + Windows 10 = best processor ever
* Mobile Xeon workstation
* Kabylake and Cannonlake on track for 16 and 17
* 50% growth of 2 in 1 in '16; it drives laptop growth
* PC: NUC, Compute Stick, gaming

Nothing new...

 

[carop]

If you accept those numbers for Samsung 20 nm, then it's not any denser than Intel's 22 nm node. (Those metal and gate pitches are exactly the same as Intel's 22 nm.) They did have a published 20 nm process with 64 nm metal pitch...I don't know if any products were released with this.

Intel N22:
Gate Pitch: 90nm
Metal1: 90nm
Metal2: 80nm
1D layouts

Samsung N20:
Gate Pitch: 90nm
Metal1: 80nm
Metal2: 80nm
2D layouts

You are suggesting that 1D layouts are density neutral. Do you have any source on this?

Thanks.

PS: The numbers are from Dick James, Chipworks:

http://www.avsusergroups.org/jtg_pdfs/JTG2015_7JamesRevSm.pdf

 

[frozentundra123456]

Getting sick of the "best processor ever" line. I wish it were a little bit "more bester", i.e. more improved relative to previous generations. The fact that they dont give any numbers, ie % lower power use or % performance improvement is telling.

 

[Arachnotronic]

Intel N22:
Gate Pitch: 90nm
Metal1: 90nm
Metal2: 80nm
1D layouts

Samsung N20:
Gate Pitch: 90nm
Metal1: 80nm
Metal2: 80nm
2D layouts

You are suggesting that 1D layouts are density neutral. Do you have any source on this?

Thanks.

PS: The numbers are from Dick James, Chipworks:

http://www.avsusergroups.org/jtg_pdfs/JTG2015_7JamesRevSm.pdf

Like I said, that Samsung 20nm node was LPE; 20nm LPP featured a 90nm gate pitch and minimum metal of 64nm. It never made it into production, probably because it wanted to focus on getting 14nm out to market ASAP.

 

[Arachnotronic]

So I haven't spent much time viewing this presentation yet, but it looks like 14nm yields are still not great and the company is significantly scaling back its investments in the mobile chip market.

Looking forward to viewing the whole presentation later today once replay is available.

 

[Systems analyst]

Progress: 14nm update + cost per transistor


Competitiveness
* Chart from last year (year of production on x-axis; transistor x metal pitch on y-axis)
* Chart from 2013! Adjusted for both SS and TSMC
* Comparison of Apple products A8 and A9, both SS and TSMC!!!! LOL 😀.
* A9 hardly denser than A8!
* BDW and SKL worse than A8 and A9! (Like 0.7x as dense) (Note: this is before normalization, so don't draw your conclusions yet 😀 😀 😀)
=> composition of die: most is SRAM, then reg files, logic short cell, IO and tall cell logic
* Comparison of compositions (in terms of % of full die for each of the 4 products)
* Normalized (for die composition) comparison: BDW and SKL 1.4x better than SS A9
* Updated comparison of 2013 chart! SS slightly better than TSMC, but still worse than Intel by a lot, basically the slide doesn't change even with the real product numbers from Apple A8/A8 substituted
* Update on 10nm numbers with available information => lead will continue (competition's 10nm will be sligtly below Intel 14nm)



Summary: 14nm maturing, cost/transistor, economics = solid, view of competition unchanged, research pipeline = full

Intel's comparison of density of transistors in Apple's A9 and Intel's Broadwell and Skylake may be flawed.
They normalise Apple's A9 density, but they ignore the fact that Intel's CPUs need a 'chipset' chip, which is on a lower density process; if they normalised taking the chipset into account, they might find a different result, I suspect.

They also seem to admit that their client group is in trouble. They do not seem to be making progress in tablets or phones, and the 2015 client group profits will be significantly less than those for PCs in 2014, to the tune of $6.6 billion, by my estimate; even if 3 - 4 billion is attributable to mobile phones and tablets, their client group is doing badly, despite their supposedly better production process.

I suspect that in five years time, most mobile computing, including laptops, will use SoCs, not CPUs, from sources other than Intel.

 

[krumme]

Intel's comparison of density of transistors in Apple's A9 and Intel's Broadwell and Skylake may be flawed.
They normalise Apple's A9 density, but they ignore the fact that Intel's CPUs need a 'chipset' chip, which is on a lower density process; if they normalised taking the chipset into account, they might find a different result, I suspect.

They also seem to admit that their client group is in trouble. They do not seem to be making progress in tablets or phones, and the 2015 client group profits will be significantly less than those for PCs in 2014, to the tune of $6.6 billion, by my estimate; even if 3 - 4 billion is attributable to mobile phones and tablets, their client group is doing badly, despite their supposedly better production process.

I suspect that in five years time, most mobile computing, including laptops, will use SoCs, not CPUs, from sources other than Intel.

Thats why they invest in areas where there is profit - as the chairman said. Its just the positive way to frame it 🙂 there is plenty opportunities for improvements. And easy ones. They just have to let it go and get on.

 

[mrmt]

So I haven't spent much time viewing this presentation yet, but it looks like 14nm yields are still not great and the company is significantly scaling back its investments in the mobile chip market.

14nm is shaping up to be a train wreck of huge proportions for Intel, and given how big 10nm leverages on it, that doesn't look good at all.

 

[Arachnotronic]

14nm is shaping up to be a train wreck of huge proportions for Intel, and given how big 10nm leverages on it, that doesn't look good at all.

Yep, notice this slide:

14nm product costs are going up for performance (i.e. large dies) and value (where they are beefing up Atom) because these segments will see the shift to 14nm next year; the mainstream costs saw a big jump this year because we saw a transition to 14nm Broadwell/Skylake here. They are also forecasting DCG op profit to grow slower than revenue because of the mix-in of 14nm Broadwell-EP/EX.

This is a mess.

 

[jpiniero]

14nm is shaping up to be a train wreck of huge proportions for Intel, and given how big 10nm leverages on it, that doesn't look good at all.

All the foundries are having problems, it's not just Intel.

 

[Khato]

Intel's comparison of density of transistors in Apple's A9 and Intel's Broadwell and Skylake may be flawed.
They normalise Apple's A9 density, but they ignore the fact that Intel's CPUs need a 'chipset' chip, which is on a lower density process; if they normalised taking the chipset into account, they might find a different result, I suspect.

How exactly would that make the comparison 'flawed' though? Given that it's already accounting for the difference between transistors, including those of the I/O variety. Intel's perfectly capable of producing a full SoC on leading edge process, it just doesn't make sense to do so in all markets.

One interesting point from that comparison though is the marked reduction in I/O between BDW and SKL. Not certain what the cause for that might be?

 

[Thala]

That density normalization argument really backfires. It essentially shows that despite having inherently higher density at process level, that the actual product having less density due to sub-optimal architecture. I bet these slides are provided by process technology to prove "Hey, it is not our fault".

 

[mrmt]

All the foundries are having problems, it's not just Intel.

Yes, but Intel problems are not the same of the other foundries.

14nm was supposed to enable two things, first a cost advantage enough to tilt the balance on mobile market towards Intel product, and a cost advantage enough to give Intel foundry business a good value proposition so it could begin to be noticed on the market. It failed miserably on these two counts, and it seems that it is also being detrimental to Intel other businesses as well.

And with 10nm being a straight shrink of this mess, that basically kills whatever chances Intel had of being a dominant mobile player, and that postpones the foundry movement towards the end of the decade, when TSMC will be even bigger than today.

I may be well be wrong and 10nm turns out to be the node that will deliver these things, but I'm not optimistic here, really. 10nm is already delayed and we are hearing about 14nm issues by the time we were supposed to be discussing its phase out. 10nm is high risk as well.