Interesting notes from the Intel Q2 2018 Call

[Daniel Nenni]

The first thing you will notice is the executive change from the previous call:

Mark H. Henninger - Intel Corp.
Brian M. Krzanich - Intel Corp.
Robert Holmes Swan - Intel Corp.

Mark H. Henninger - Intel Corp.
Robert Holmes Swan - Intel Corp.
Venkata S. M. Renduchintala - Intel Corp.
Navin Shenoy - Intel Corp.


Navin is VP/GM of the Data Center group and Murthy (Venkata) has a handful titles so nobody really knows what he actually does. Intel posted some impressive numbers but the call itself was typical Intel doublespeak:

• Record second-quarter revenue was $17.0 billion, up 15 percent year-over-year (YoY); data-centricbusinesses* grew 26 percent and PC-centric revenue grew 6 percent.
  
• GAAP earnings-per-share (EPS) of $1.05 rose 82 percent YoY; non-GAAP EPS of $1.04 was up 44 percent.
  
• Year-to-date, generated $13.7 billion in cash from operations, $6.3 billion of free cash flow and returned $8.6billion to shareholders (dividends of $2.8 billion and share repurchases of $5.8 billion).
  
• Raising full-year revenue outlook to approximately $69.5 billion, GAAP EPS outlook to approximately $4.10 andnon-GAAP EPS of $4.15; up $2.0 billion, $0.31 and $0.30 from April guidance, respectively.

The big interest was 10nm and why it is delayed. Questions were asked, Murthy answered them in Intel doublespeak like he had worked there for his entire career. I was hoping for a much more technical answer but then I remember that Murthy has a BSEE followed by an MBA and a PhD in Digital Communications.


10nm Question:


Venkata S. M. Renduchintala - Intel Corp.
Hi, this is Murthy. I'll take that one. We continue to make progress on 10-nanometer. Yields are improving consistent with the timelines we shared in April. And yes, you're quite right. The systems on shelves that we expect in holiday 2019 will be client systems, with data center products to follow shortly after.


10nm question:


Venkata S. M. Renduchintala - Intel Corp.
So let me take that. I think as we look at what we need to do in 10-nanometers, again, let me replay some of the data we shared on our April call. Recall that 10-nanometers strive for a very aggressive density improvement target beyond 14-nanometers, almost 2.7x scaling. And really, the challenges that we're facing on 10-nanometers is delivering on all the revolutionary modules that ultimately deliver on that program. And while there's risk and a degree of delay in our timeline on that, we're very pleased with the resiliency of our 14-nanometer roadmap, where in the last few years we've delivered in excess of 70% product performance improvement as we've moved through our 14-nanometer generation of products.

So as we look at 2019 across both the client and data center space, we feel very good about the product competitiveness of our 14-nanometer program, and that to some degree is factoring into our timing on 10-nanometer and launching 10-nanometer at a point in time where we believe the yields are at a level that make it prime for volume production. So 14-nanometer I think through the rest of this year and through 2019 continues, we believe, to drive product leadership across all our portfolio in client and server.


7nm Question:


Venkata S. M. Renduchintala - Intel Corp.
Sure. Blayne, so 7-nanometer is very much R&D in deep progress, and we're making good progress on that development. We're not giving a direct timeline right now. But we've also made some fairly judicious choices in defining 7-nanometer, learning from our 10-nanometer experiences. And we're focusing on an optimum balance point between density, power and performance, and schedule predictability. So I think what you'll see is a more balanced approach across those three vectors.

So we're still going to drive density but balancing that with a continued focus on driving transistor performance at the same time, which is highly valued as ASP drivers both in our client and server businesses. And we're really also focusing on being much more precise in our ability to launch. So those are the key learnings that are coming out of 10-nanometer as we go into 7-nanometer. And as we monitor progress on 7-nanometer just as closely as we are on 10-nanometer, I feel those lessons are being well absorbed into our progress, and we're lining up to support our product plans as our roadmap dictates.


10nm Question:


Venkata S. M. Renduchintala - Intel Corp.
In general, we're going to see a much shorter ramp period between our products going forward in client and server. So yes, I think it's a good observation that as we talk about client systems on shelf by the second half of 2018, you shouldn't expect too much of a delay before you see data center products coming out.

So much closer proximity, albeit slightly delayed than we expected on 10-nanometer, and then you should see that pretty much improve to almost parity of launches as we get into later technologies. So the traditional model of server following rather lengthily after client is probably going to become more sequential going forward.


"Client systems on the shelf by the second half of 2018" ?!?!?


10nm Question:


Venkata S. M. Renduchintala - Intel Corp.
I just wanted to add to Bob's perspective. The way I look at our roadmap, what we're really focused on is delivering product leadership generation after generation, and that's the system level. And while processes are a very important part of that recipe, so are the other ingredients as well, such as product architecture, silicon design, and packaging.

And to Bob's point, as we look towards our roadmap, gross margin maximization is going to come from delivering excellence that drives performance, and therefore we're taking a very balanced view in all the ingredients that go into that. So you should expect in our roadmap going forward a much longer overlap between generations of technology as we try and make sure that, along with process, we add the other ingredients of technology leadership that will become more and more apparent.


If this was a CEO tryout I would say that Murthy passed the Intel doublespeak test with flying colors. And last but not least the fitting sendoff of BK who the Intel BoD completely threw under the bus:


Robert Holmes Swan - Intel Corp.
Personally and on behalf of Intel's 100,000-plus employees, I'd like to thank Brian [Krzanich] for his many contributions to the company over his 35-year career. The investments he made set us on a course for transformation. Even more importantly, he developed the right strategy and leadership team to carry that transformation forward while we conduct the CEO search. Our financial results in the second quarter show we're doing just that.

 

[Jozo035]

AMD said that they are already sampling 7nm ROME server CPUs made in TSMC with release planned early next year.

Does it really mean that they will have 1 year advantage over Intel? Because it would be really big.

And regarding 7nm, they planned another big density jump, probably this time without performance hit (as seen at 10nm), EUV (or maybe SADP EUV), new transistor architecture (to don't fall behind foundries) and probably step forward with materials... All this at single node transition and they are sure that it will be easier than 10nm? Well, i don't have info about progress they made but for me it looks like that 7nm is even more challenging than 10nm.

 

[Daniel Nenni]

AMD said that they are already sampling 7nm ROME server CPUs made in TSMC with release planned early next year.

Does it really mean that they will have 1 year advantage over Intel? Because it would be really big.

No it does not. Remember, Intel is talking about product on the shelf while TSMC and GF are talking about wafers and AMD is talking about chips. It takes Apple a good six months to get from wafer to product on shelf and they develop their own silicon. I'm guessing it takes the Intel client/PC customers a bit longer.

To me it looks like both Intel and AMD will have product on the shelf at about the same time which is great for AMD who is usually behind. I do think however that Intel 10nm silicon will again beat the foundries 7nm in regards to density and performance. The architecture of the AMD chips may make up for that but process to process comparisons will favor Intel.

 

[Daniel Nenni]

And regarding 7nm, they planned another big density jump, probably this time without performance hit (as seen at 10nm), EUV (or maybe SADP EUV), new transistor architecture (to don't fall behind foundries) and probably step forward with materials... All this at single node transition and they are sure that it will be easier than 10nm? Well, i don't have info about progress they made but for me it looks like that 7nm is even more challenging than 10nm.

EUV will reduce the complexity of design and manufacturing by eliminating multi patterning so 7nm should be less challenging in theory. The challenge is that the EUV equipment is not yet production worthy. Hopefully that will be resolved in time for 5nm (Intel 7nm).

It really is interesting to compare the Intel and TSMC conference calls. When TSMC does not want to disclose information they just say so. Intel on the other hand responds with double speak. Do they really think the analysts and industry on a whole does not know this? Intel culture must change and hiring a CEO from within will not cut it, my opinion.

 

[IanD]

No it does not. Remember, Intel is talking about product on the shelf while TSMC and GF are talking about wafers and AMD is talking about chips. It takes Apple a good six months to get from wafer to product on shelf and they develop their own silicon. I'm guessing it takes the Intel client/PC customers a bit longer.

To me it looks like both Intel and AMD will have product on the shelf at about the same time which is great for AMD who is usually behind. I do think however that Intel 10nm silicon will again beat the foundries 7nm in regards to density and performance. The architecture of the AMD chips may make up for that but process to process comparisons will favor Intel.

That doesn't stack up with Intel's own roadmap showing Ice lake (10nm server part) due to be launched in 2Q2020, with "customer availability" 1-2 quarters later -- so presumably 3Q2020 for mass production. AMD Rome is sampling now and planned to be in production in 1Q2019, which I make to be about 18 months ahead of Ice Lake which it was designed to compete with -- this is a huge threat to Intel's lucrative server business.

Also there's no evidence that Intel 10nm will beat foundry 7nm on density and performance -- and given that it will be way more than 1 year later to market, any small differences will be swamped by this delay, it'll be competing with the 2nd generation 7nm processes by then (like 7FF+ using EUV) which are likely to be slightly ahead of Intel 10nm.

 

[count]

No it does not. Remember, Intel is talking about product on the shelf while TSMC and GF are talking about wafers and AMD is talking about chips. It takes Apple a good six months to get from wafer to product on shelf and they develop their own silicon. I'm guessing it takes the Intel client/PC customers a bit longer.

To me it looks like both Intel and AMD will have product on the shelf at about the same time which is great for AMD who is usually behind. I do think however that Intel 10nm silicon will again beat the foundries 7nm in regards to density and performance. The architecture of the AMD chips may make up for that but process to process comparisons will favor Intel.

I don't trust Intel anymore when they start talking timelines. They say they are shipping 10nm now, when the reality is they are shipping crippled parts to Chinese vendors in order to make that claim. Who knows what they mean when they say they will be shipping server parts in 2019.

 

[Chipper]

I think they expect competitive price pressure in Q4 2018 from somewhere. They are guiding to 59% Q4 gross margins from four factors; modem mix up, NSG up, 10nm costs up and the number four that Swan added without being prompted: "and lastly, just our outlook for ASPs. ASP has been a significant contributor to our high gross margin performance over the last couple of quarters. Now, we've assumed that it won't stay at those levels." Ouch.

This also has to be the first call where "Moore's Law" was not mentioned. Instead it was talk about increased process complexity, more reliance on older nodes, how great it is to have a broad product line and "balancing" node designs with manufacturability. My question is the extent to which their unique architecture limits their node migration, in addition to cobalt use. Both TSM and AMD called out "architecture" as a competitive advantage at 7nm. Imagine; INTC management extolling the virtues of their modem business. How far the mighty have fallen.

 

[Daniel Nenni]

I don't trust Intel anymore when they start talking timelines. They say they are shipping 10nm now, when the reality is they are shipping crippled parts to Chinese vendors in order to make that claim. Who knows what they mean when they say they will be shipping server parts in 2019.

I agree, I do not believe anything that comes out of Intel Corporate. The 10nm parts that Intel ships are from the OR R&D fab. They may not be crippled, just low yielding, at at least I hope so. If yield really is the issue, which I believe to be true, Intel can unleash 10nm at anytime they want by sacrificing margin. To me sacrificing margin is a whole lot better than losing market share to AMD so either Intel Corporate knows something we don't or their are just being their arrogant selves.

View attachment 22113What happened to "Only the Paranoid Survive"?!?!?!?!

"How can you motivate yourself to continue to follow a leader when he appears to be going around in circles?"

"Businesses fail either because they leave their customers or because their customers leave them!"

"In technology, whatever can be done will be done."

"People in the trenches are usually in touch with impending changes early."

Someone needs to smack Andy Bryant with this book a few times!

 

[Chipper]

They mentioned EUV in a confusing way, noting: "with the exception of particularly 10nm going to 7nm, litho changing quite a bit with EUV, with each node, even with more capital intensity, we have relatively high reuse and we can leverage the tools for longer." Does that mean they introduce EUV after this node? How many layers? Do they only have one machine? If TSM buys most of the machines for the balance of this year, where does that leave their ramp?

INTC management also bragged about how they have improved 14nm performance 70% in an undetermined time frame, but possibly five years. Some deviation from Moore's Law doubling every 12 months, no 18 months, no 24 months, no 70% in 60 months.

Also their risk factors are a total haha: security vulnerabilities and their remediation reducing the competitiveness of their products and adversely impacting customer relationships. They also noted how they could be subject to litigation involving among the usual items "disclosure and other issues", like that October stock sale by BK? This is a legal document that needs to be more forthcoming than their conference calls so it is a better read into their real thinking. The whole security paragraph was added in Q1 so it is a new concern not discussed in the call but real none-the-less.

 

[hkwint]

Murthy's TL;DR 10nm reply

I agree, I do not believe anything that comes out of Intel Corporate. The 10nm parts that Intel ships are from the OR R&D fab. They may not be crippled, just low yielding, at at least I hope so. If yield really is the issue, which I believe to be true, Intel can unleash 10nm at anytime they want by sacrificing margin.

I consider myself luck, transcripts are available for free this time!

Edited Transcript of INTC earnings conference call or presentation 26-Jul-18 9:00pm GMT

Pierre C. Ferragu, New Street Research LLP ...: And I have 2 questions ... The first one is, could you give us some color on what are the most challenging aspects of the process (10nm - hk) that you need to address to improve yields? ...

...Murthy...:

(hk - yada yada 14nm great yada yada) ... And that, .... is factoring into our timing of 10-nanometer and launching 10-nanometer at a point in time where we believe the yields are at a level that make it prime for volume production. (emphasis: hk) So 14-nanometer, I think through the rest of this year and through 2019, continues we believe to drive product leadership across all our portfolios in clients and server.

--> If 14nm drives product leadership through 2019, then that implies 10nm doesn't, or am I being crazy here???

 

[Fred Chen]

Intel 7nm might be past 2021, so they must be considering the High-NA EUV.

 

[raghu78]

No it does not. Remember, Intel is talking about product on the shelf while TSMC and GF are talking about wafers and AMD is talking about chips. It takes Apple a good six months to get from wafer to product on shelf and they develop their own silicon. I'm guessing it takes the Intel client/PC customers a bit longer.

To me it looks like both Intel and AMD will have product on the shelf at about the same time which is great for AMD who is usually behind. I do think however that Intel 10nm silicon will again beat the foundries 7nm in regards to density and performance. The architecture of the AMD chips may make up for that but process to process comparisons will favor Intel.

Daniel I have to disagree with you on this one. Intel saying 10nm systems in shelves for holiday 2019 is typical PR talk. At this point Intel has no credibility left. They have been saying 10nm will be ready from late 2016 and now 10nm is a 2020 affair. By that time TSMC would be getting ready for N5. Intel went from 1 year ahead at 14nm vs TSMC 16nm to 2 years behind at 10nm vs TSMC 7nm. You need to go back and look at Broadwell-Y launch back in Nov 2014 . It took 9-10 months for Skylake to launch in real volume in Aug 2015 across desktop and notebooks. I think we will something similar with ICL-Y launching in tablets and then ICL-S arriving by mid 2020. ICL-SP is now almost definitely not happening in 2020. Intel's recent HPC roadmap shows Cascade Lake SP launch in H1 2019 and Cooper Lake SP for mid 2020. Why would Intel bring Cooper Lake SP if ICL-SP is scheduled for mid 2020. At this point I have to say Charlie's sources are spot on and Intel is in full on panic mode.

btw Rome is looking good to launch in Q2 2019. Rome has probably taped out in Q1 2018 and AMD has had first silicon back for a while and is now sampling server customers so that they can do their qualification in parallel to AMD's validation and debugging of Rome silicon. I am quite sure we will see serious volume from AMD Rome in H2 2019. For Intel I would say 10nm volume in 2019 is going to be negligible. The big problem for Intel is they are going to pit a 28c Cascade Lake SP vs 64c Rome in H2 2019. Whats even more damning is 2020 when 14++ Cooper Lake SP will go up against Milan at TSMC N7+ . I think Intel is going to have a few very bad years before they recover from the 10nm debacle.

I agree, I do not believe anything that comes out of Intel Corporate. The 10nm parts that Intel ships are from the OR R&D fab. They may not be crippled, just low yielding, at at least I hope so. If yield really is the issue, which I believe to be true, Intel can unleash 10nm at anytime they want by sacrificing margin. To me sacrificing margin is a whole lot better than losing market share to AMD so either Intel Corporate knows something we don't or their are just being their arrogant selves.

View attachment 22113What happened to "Only the Paranoid Survive"?!?!?!?!

"How can you motivate yourself to continue to follow a leader when he appears to be going around in circles?"

"Businesses fail either because they leave their customers or because their customers leave them!"

"In technology, whatever can be done will be done."

"People in the trenches are usually in touch with impending changes early."

Someone needs to smack Andy Bryant with this book a few times!

Andy Bryant needs to go. In fact Intel's BoD needs a complete overhaul. The fact that they picked somebody like Brian Krzanich who was completely unsuited for the CEO role says a lot. Daniel you are assuming that Intel 10nm has only yield issues. If you look at Cannonlake chip launched it has the entire iGPU disabled and it has 2C/4T for 15w TDP. Its a really bad part if you look at what a full node shrink is supposed to bring.

https://ark.intel.com/products/136863/Intel-Core-i3-8121U-Processor-4M-Cache-up-to-3_20-GHz

Look at what Intel is shipping on 14+ KBL-R for the same 15w TDP. the core i7 8650u has 4C/8T with a fully enabled iGPU for 15w TDP.

https://ark.intel.com/products/124968/Intel-Core-i7-8650U-Processor-8M-Cache-up-to-4_20-GHz

The 8121 is an economically unfeasible product on a broken process. Not hitting expected perf targets is one thing but when you have to disable half the chip and still your TDP rating is roughly twice vs last gen product built on last gen node (when looking at number of cores) its just ugly. I don't know how Intel can ship that part and say they are in volume production with a straight face.

 

[raghu78]

That doesn't stack up with Intel's own roadmap showing Ice lake (10nm server part) due to be launched in 2Q2020, with "customer availability" 1-2 quarters later -- so presumably 3Q2020 for mass production. AMD Rome is sampling now and planned to be in production in 1Q2019, which I make to be about 18 months ahead of Ice Lake which it was designed to compete with -- this is a huge threat to Intel's lucrative server business.

Also there's no evidence that Intel 10nm will beat foundry 7nm on density and performance -- and given that it will be way more than 1 year later to market, any small differences will be swamped by this delay, it'll be competing with the 2nd generation 7nm processes by then (like 7FF+ using EUV) which are likely to be slightly ahead of Intel 10nm.

Yeah Rome looks to be on target for a launch in Q2 2019 atleast. I think Rome on TSMC N7 will compete with 14++ Cascade Lake SP and Milan on TSMC N7+ with EUV will compete with 14++ Cooper Lake SP. Charlie wrote an article about Cooper Lake specs with an AP version for higher core counts to compete with 64c/128t Rome. The TDP rating for Cooper Lake AP is simply ridiculous.

 

[Fred Chen]

Also there's no evidence that Intel 10nm will beat foundry 7nm on density and performance -- and given that it will be way more than 1 year later to market, any small differences will be swamped by this delay, it'll be competing with the 2nd generation 7nm processes by then (like 7FF+ using EUV) which are likely to be slightly ahead of Intel 10nm.

Intel seemed to have admitted going from 14nm to 10nm was too much of a density jump, it suggests they will dial back their 10nm to the current foundry 7nm density. That might also help their performance.

 

[benb]

Biggest disappointment with this conference call was no CEO announcement.

They are in an existential crisis with no CEO.

This is not a continue forward with minor tweaks situation. They need to sell stuff, close stuff, get smaller, and focus on manufacturing.

Leadership in semiconductors passes every 10-20 years. NEC passed to IBM, then IBM to Motorola, then Motorola to Intel. Intel passed to Samsung last year.

Portland TD is one of the things that needs to close. A viable 7nm process will need to be purchased. Future TD should not be in Portland; I would suggest Israel. US based TD is the core problem. Intel, being global, luckily, has other promising options.

 

[mbello]

Intel's financial performance is impressive, 15% revenue growth YoY is great.
If Daniel is right, Intel will now throw a lot of money on their execution problem, dial ASP down a bit, add some extra goodies on their chips and hope to fence AMD off as they always did. If they succeed, they will continue delivering good news to Wall Street and soon enough their 10nm failure will be behind them.

On the other hand, AMD is taking every chance against Intel and going TSMC may end up being key to them. Nobody here spoke about how TSMC is seeing this opportunity. Are they treating AMD just like every other client or is TSMC seeing AMD as their chance to eat Intel's lunch?

What if TSMC itself decides to throw money on their AMD relationship? TSMC may be too reliant on Apple nowadays, and what better client than AMD to drive future growth? Imagine if AMD succeeds and reaches 33% market share for server chips 3 years from now, what could that mean to TSMC? And TSMC knows that it can enable AMD to compete with Intel like it never did before.

I believe both AMD and TSMC have a huge opportunity in front of them, just do not know if TSMC is in fact betting high on it. Any clues?

 

[benb]

What we know for sure is Intel engineering, once a real strength, is failing to deliver the goods. We don't know yet if AMD (or TSMC, or Global) can execute and deliver superior value. There is so much exuberance surrounding AMD at the moment, probably too much (akin to Tesla and the Model 3, leaping past all the execution risks, spinning stories).

A less exuberant take would be to expect Intel to push their 14nm advantages while AMD, TSMC and Global have some struggles with 7nm. Clock speed and core counts still seem to tip in Intel's favor, today. Intel can cut their margins and undercut AMD prices.

AMD has wisely chosen 2 partners. I think TSMC fills a need for supply assurance, having more than one fab. TSMC can't be overly aggressive with their AMD commitments because they are wise (not exuberant) and because they are in a split business.

 

[Daniel Nenni]

Intel's financial performance is impressive, 15% revenue growth YoY is great.
If Daniel is right, Intel will now throw a lot of money on their execution problem, dial ASP down a bit, add some extra goodies on their chips and hope to fence AMD off as they always did. If they succeed, they will continue delivering good news to Wall Street and soon enough their 10nm failure will be behind them.

On the other hand, AMD is taking every chance against Intel and going TSMC may end up being key to them. Nobody here spoke about how TSMC is seeing this opportunity. Are they treating AMD just like every other client or is TSMC seeing AMD as their chance to eat Intel's lunch?

What if TSMC itself decides to throw money on their AMD relationship? TSMC may be too reliant on Apple nowadays, and what better client than AMD to drive future growth? Imagine if AMD succeeds and reaches 33% market share for server chips 3 years from now, what could that mean to TSMC? And TSMC knows that it can enable AMD to compete with Intel like it never did before.

I believe both AMD and TSMC have a huge opportunity in front of them, just do not know if TSMC is in fact betting high on it. Any clues?

GF designed 7nm specifically for AMD (HP) like TSMC does for Apple. It is my hope that the AMD parts on GF 7nm will perform well.

The TSMC HP process is developed with/for NVIDIA. Since AMD and NVIDIA compete on the GPU side it will be interesting to see if AMD is invited to the TSMC inner circle at 5nm. The real question is: Can AMD write some VERY big TSMC checks? Last I looked at their balance sheet they cannot.

 

[Daniel Nenni]

Biggest disappointment with this conference call was no CEO announcement.

They are in an existential crisis with no CEO.

This is not a continue forward with minor tweaks situation. They need to sell stuff, close stuff, get smaller, and focus on manufacturing.

Leadership in semiconductors passes every 10-20 years. NEC passed to IBM, then IBM to Motorola, then Motorola to Intel. Intel passed to Samsung last year.

Portland TD is one of the things that needs to close. A viable 7nm process will need to be purchased. Future TD should not be in Portland; I would suggest Israel. US based TD is the core problem. Intel, being global, luckily, has other promising options.

I agree completely. Rather than throwing BK under the bus and hurting his family why didn't Andy Bryant and the Intel BoD pick a successor and smoothly transition BK out? Seriously, what was the rush? Hopefully there is much more to this story otherwise the Intel Bod is brain dead, my opinion.

I did hear from a reputable source that Diane Bryant is not in the running. Apparently she did not leave on good terms. When BK was hired I strongly urged Intel to acquire NVIDIA to fill the CEO slot. That would have been a very disrupted move. Today I strongly urge Intel to hire from the outside. If Intel promotes Murthy or one of the other Intel insiders I predict some very mediocre times ahead for Intel, absolutely.

 

[benb]

It makes me very sad, what’s happening, and I guess I’m not the only one (“So it goes”): Intel and the Danger of Integration – Stratechery by Ben Thompson

An outsider CEO could catalyze a change in culture. I think sites need to close too. This guy is writing about the toxic culture at Intel:

https://mondaynote.com/intels-toxic-culture-1b79905adf45

I think it’s telling that the stock market didn’t sell off on Facebook or Netflix earnings misses, but Intels. What happens next at Intel matters to everyone.