Intel Investor Meeting 2017: February 9

[dark zero]

Sure they do. Its about growth factors. Its a reshuffle like always. Desktop lost to mobile previously as well.

The ARM server dreams is nothing but a pipedream. Company after company tries, company after company abandons it.

So is Intel approach to mobile. They failed with SOFIA that killed along with it the only true competitor of ARM.

 

[Arachnotronic]

So let me get this straight:

10 nm: Cannonlake (U/Y only) and Coffee Lake on 14 nm
10 nm+: Cannonlake-X (but no EP) then Icelake
10 nm++: Icelake-EP/X then Tigerlake
7 nm: Sapphire Rapid?

I don't know if Cannon Lake-EP/EX/X exists, but Diane said DCG would take the lead in process at 10nm++.

So I would expect Ice Lake mainstream to be first products on 10nm+, then Ice Lake-EP/EX to be the first product on 10nm++, followed by Tiger Lake. Then Sapphire Rapid on 7nm.

 

[raghu78]

Arachnotronic a simple question. Intel has been moving to a new process node with < 100 sq mm dies and 5w-15w chips at 14nm and 10nm. Next year we will have 14nm++ and 10nm coexisting. I see a similar situation with 10nm++ and 7nm . So when Intel say that DCG will move first to 10nm++ and 7nm first what do they mean ? How can something be manufactured at 2 nodes which will pretty much overlap in time ?

 

[JDG1980]

They aren't going to need to yield massive dies. They are going to "chop up" the dies and stitch them back together using EMIB packaging. Intel says that it will be virtually indistinguishable from a monolithic die, none of the issues associated with traditional MCM.

This sounds very similar to what we've heard about AMD's upcoming "Infinity Fabric".

 

[JDG1980]

Cloud is as locked to x86 as much as the Enterprise. There isn't anything else that will replace x86. And no, ARM is a complete joke in the area.

I think you overestimate the level of x86 lock-in on servers. Enterprise workstations are stuck on x86 for a variety of reasons; many of them have mission critical software in the form of binary blobs written long ago, for which the source code isn't even available and wouldn't be portable if it was. But servers are far more likely to run software that the maintainers have the ability to recompile. The biggest companies that run full in-house stacks (Google, Facebook, etc) would go to ARM in a minute if the TCO justified it.

But even if you're right that servers will be "locked to x86", that still doesn't get Intel off the hook with competition from AMD's Zen.

 

[Arachnotronic]

Arachnotronic a simple question. Intel has been moving to a new process node with < 100 sq mm dies and 5w-15w chips at 14nm and 10nm. Next year we will have 14nm++ and 10nm coexisting. I see a similar situation with 10nm++ and 7nm . So when Intel say that DCG will move first to 10nm++ and 7nm first what do they mean ? How can something be manufactured at 2 nodes which will pretty much overlap in time ?

Here is my understanding/guesses, this is not official info, so it comes with no guarantee what-so-ever that it's right 😉

Late 2017-early 2018: Cannon Lake-Y on 10nm, everything else on 14nm++ (Coffee Lake-U/S/H, Kaby Lake Refresh U).

Late 2018: Begin manufacturing Ice Lake-Y/S/H/U on 10nm+. I would expect some Ice Lake product in the market during 2H 2018, but with the full launch like we saw with Kaby in very early 2019 (hopefully I'm wrong and we get the entire stack in late 2018).

Late 2019: Begin manufacturing Tiger Lake on 10nm++ -- the "third wave" of 10nm -- for client machines. Again, I expect some products in market in late 2019, with wider SKU availability in early 2020.

If Tiger Lake client trickles into market in late 2019, with bigger volume in early 2020, then I would imagine that you'll see Ice Lake-EP/EX in the market perhaps in mid-2019. Then, we'll see the first 7nm server family, Sapphire Rapid, in mid-2020.

 

[Arachnotronic]

This sounds very similar to what we've heard about AMD's upcoming "Infinity Fabric".

AMD's Naples is still a typical MCM, not quite what's going on here with the "chop it up and connect it via EMIB."

The nice thing about EMIB is that you are not limited to the size of the interposer because it's just silicon between the "connecting" dies -- in other words, it just needs to cover the parts that connect. EMIB can also be used as a more cost effective way to include technologies like HBM2/HMC/etc. so don't be surprised if future Intel Xeon processors use HMC or HBM2/3.

 

[NTMBK]

The EMIB stuff is quite exciting. It gives a lot of the benefits AMD talked about in their "Die Stacking Is Happening" talk, but without the cost and size limits of a chip-sized interposer. You could see chips made on multiple different specialised process variants (one targeted at GPU density, one going for CPU frequency) or totally different processes (like a DRAM process, or NVRAM) all combined into one.

 

[raghu78]

Arachnotronic let me understand what you say.

Late 2017/early 2018 - 10nm Cannonlake
Late 2018/early 2019 - 10nm+ Icelake client
Mid 2019 - 10nm+ Icelake EP / EX
Late 2019 - 10nm++ Tigerlake client
Mid 2020 - 7nm Sapphire Rapid server.

You are directly contradicting what Intel says. They say server will be the first to 10nm++ and 7nm. Moreover you are saying Intel will have 10nm server chips out for 1 year before they replace with 7nm. It defies logic. Have you seen how long Intel server chips have stayed on a process node at 22nm and 14nm. Ivy Ridge EP - Sep 2013 Broadwell EP - Q2 2016 .Icelake EP - Q2 2019 (i think its going to be later but lets go with best case) . 2.5 - 3 years is the norm. You are saying 12 months. Moreover EX historically is a year behind EP due to the larger die sizes. Even assuming that Intel will move to EMIB at 10nm++ its very difficult to see EP and EX launch at same time in Q2 2019. This is the problem with Intel's roadmap. Their plans for node iterations and architecture iterations is vague and does not give much confidence given how poorly they are executing with 10nm. If 10nm was in good shape we should be seeing a strong ramp in H2 2017 for Cannonlake with a tock by late Q3 2018 just like we saw with Skylake in Q3 2015.

 

[mikk]

Late 2017-early 2018: Cannon Lake-Y on 10nm, everything else on 14nm++ (Coffee Lake-U/S/H, Kaby Lake Refresh U).

Cannonlake-U is on 10nm as well.

 

[Arachnotronic]

Arachnotronic let me understand what you say.

Late 2017/early 2018 - 10nm Cannonlake
Late 2018/early 2019 - 10nm+ Icelake client
Mid 2019 - 10nm+ Icelake EP / EX
Late 2019 - 10nm++ Tigerlake client
Mid 2020 - 7nm Sapphire Rapid server.

You are directly contradicting what Intel says. They say server will be the first to 10nm++ and 7nm. Moreover you are saying Intel will have 10nm server chips out for 1 year before they replace with 7nm. It defies logic. Have you seen how long Intel server chips have stayed on a process node at 22nm and 14nm. Ivy Ridge EP - Sep 2013 Broadwell EP - Q2 2016 .Icelake EP - Q2 2019 (i think its going to be later but lets go with best case) . 2.5 - 3 years is the norm. You are saying 12 months. Moreover EX historically is a year behind EP due to the larger die sizes. Even assuming that Intel will move to EMIB at 10nm++ its very difficult to see EP and EX launch at same time in Q2 2019. This is the problem with Intel's roadmap. Their plans for node iterations and architecture iterations is vague and does not give much confidence given how poorly they are executing with 10nm. If 10nm was in good shape we should be seeing a strong ramp in H2 2017 for Cannonlake with a tock by late Q3 2018 just like we saw with Skylake in Q3 2015.

Ice Lake EP/EX I said would be 10nm++. Remember, Skylake-S was 14nm, but Skylake-EP/EX is 14nm+.

I don't have architectural details on Tiger, but my guess right now is that it's an "optimization" of Ice Lake, so 10nm++ implementation of Ice.

 

[Arachnotronic]

The EMIB stuff is quite exciting. It gives a lot of the benefits AMD talked about in their "Die Stacking Is Happening" talk, but without the cost and size limits of a chip-sized interposer. You could see chips made on multiple different specialised process variants (one targeted at GPU density, one going for CPU frequency) or totally different processes (like a DRAM process, or NVRAM) all combined into one.

Intel spends a truckload on R&D for packaging technology (they say 2x the leading two OSATs combined) and now it's paying off for them in a big way.

 

[witeken]

Intel spends a truckload on R&D for packaging technology (they say 2x the leading two OSATs combined) and now it's paying off for them in a big way.

If they spend a multiple of rest of industry, why don't they have inFO?

 

[Lidiya Nikolayev]

If anyone has more information about opportunities and potential problems Intel is facing, please share if there is public information to back up your statement.

 

[FIVR]

Did you see their CEO looking like a muppet next to that senile reality tv star we call a president? He is a liability. He has made consistently bad management decisions. Intel is a capable company staffed by some of the best engineers in the world with some of the worst upper management out there. And BK is no exception.

 

[itsmydamnation]

If anyone has more information about opportunities and potential problems Intel is facing, please share if there is public information to back up your statement.

Or how about you do your own research...........
mine of you shows this:

Announcement: Hello Friends!, I've decided to return to my passion of acting until l land a bank job in an area I am interested in or Finance Analyst or Investment Associate position after graduation.

Im fine with giving my opinion and why to people who want to discuss something, not going to give it to someone who turns around and tries to use my position and my logic to make a buck...........

 

[witeken]

If anyone has more information about opportunities and potential problems Intel is facing, please share if there is public information to back up your statement.

That's vague.

 

[imported_bman]

If anyone has more information about opportunities and potential problems Intel is facing, please share if there is public information to back up your statement.

Do your own homework.

 

[Lidiya Nikolayev]

Do your own homework.

I am but there's so much out there and its about gathering the right data

 

[Lidiya Nikolayev]

That's vague.

I am focusing on the Client Computing Group; mobile is practically a goner and I got enough info on them not to pursue but even though the Data Center brings in a large percentage of revenue I still want to find about other hurdles they are facing. What I am getting is mostly positive articles. From what I have researched management and marketing issues have caused a chain of problems in various areas so I would like to know about them if possible.

 

[Lidiya Nikolayev]

Or how about you do your own research...........
mine of you shows this:



Im fine with giving my opinion and why to people who want to discuss something, not going to give it to someone who turns around and tries to use my position and my logic to make a buck...........

I have been doing research for the past 2 months, but it's difficult to find as many "warning" or problem articles versus positive ones. How can you see the signs or get a better understand of why losses are occurring if you don't have a greater depth of information into it?

 

[Lidiya Nikolayev]

Or how about you do your own research...........
mine of you shows this:



Im fine with giving my opinion and why to people who want to discuss something, not going to give it to someone who turns around and tries to use my position and my logic to make a buck...........

I am not making anything out of it. I am in a cap stone course at my university and my research project is Intel. I have plenty of connections in Intel but was even provided a confidential agreement by my university but the Intel Lawyers restricted my friends from sharing information, so I have been doing research nearly none-stop for 2 months. Hence why if you tell me " this area has issues, then I can go pinpoint it more closely and search for articles regarding problems they might be facing.

 

[Lidiya Nikolayev]

I am not making anything out of it. I am in a cap stone course at my university and my research project is Intel. I have plenty of connections in Intel but was even provided a confidential agreement by my university but the Intel Lawyers restricted my friends from sharing information, so I have been doing research nearly none-stop for 2 months. Hence why if you tell me " this area has issues, then I can go pinpoint it more closely and search for articles regarding problems they might be facing.

It's more about getting potential opinions from the IT people because its not my department. I am a finance major. I can see things in financial statements but getting the story behind what's happening requires more digging.

 

[Lidiya Nikolayev]

Or how about you do your own research...........
mine of you shows this:



Im fine with giving my opinion and why to people who want to discuss something, not going to give it to someone who turns around and tries to use my position and my logic to make a buck...........

Also that quote your posting is from my professional site for my hobby , it has nothing to do with my research project . I thought IT people were friendly but apparently everyone is so antisocial that they are putting down a student for trying to dig deeper into a company for her project. Bad manners!!

 

[dealcorn]

Lidiya:

1. In a different thread I just asked a kinda dumb question and received a response that was fast, accurate, completely responsive and I acted on the information within 15 minutes. A binary model of whether IT folks are antisocial vs helpful is just the wrong model.

2. Anandtech strives to have a breadth of appeal that is comparable to Alice’s Restaurant. Still, there are well thought out and sound reasons why the babe thread is not located in the CPU’s and Overclocking forum. I immediately dismissed your initial concerns based solely on inappropriate avatar selection (subsequently withdrawn). Perhaps my biases against too much skirt are not unique.

3. Some dead guy said something relevant about leadership, H20 and a horse. Both Seeking Alpha and Motley Fool frequently refer to Anandtech as a reasonably authoritative site for often contentious information about hurdles Intel is facing. I find this site is positively littered with that kind of information. However, my formal training is in finance and tech is just a 40 year interest so maybe I do not understand. In this metaphor, I am not saying you are all wet.