Intel delays 14 nm KabyLake to 2016H2/2017Q1, 10 nm Cannonlake to 2017H2/2018Q1

[tenks]

Benchlife is the source

https://benchlife.info/intel-kaby-lake-scedule-show-10nm-cannonlake-will-push-back-to-2018-10162015/

 

[mikk]

I think everyone assumed in the first half

Nobody well informed did. Intel told H2 2017.

 

[Fjodor2001]

http://wccftech.com/intels-10nm-can...ies-q3-2016-kaby-lakes-desktop-chips-1h-2017/

Original source (BenchLife, in Chinese):
https://benchlife.info/intel-kaby-lake-scedule-show-10nm-cannonlake-will-push-back-to-2018-10162015/
Google translated to English:
https://translate.google.com/translate?sl=auto&tl=en&js=y&prev=_t&hl=en&ie=UTF-8&u=https%3A%2F%2Fbenchlife.info%2Fintel-kaby-lake-scedule-show-10nm-cannonlake-will-push-back-to-2018-10162015%2F&edit-text=

WccfTech:

Intel Cannonlake Chips Delayed till 2H 2017, Kaby Lake Fills The Gap in 2016-2017

Poor yields have been a major issues which have affected the production of chips with smaller process nodes. Intel Broadwell was the first chip that was affected due to poor yields since it was making use of a new 14nm process node and it came two years after the launch of Haswell (Haswell Desktop came in 2013 and Broadwell desktop came in 2015).
[...]

This delay didn’t affect mobility years as much as it affected desktop users as Intel themselves revealed that it was a mistake to leave the desktop PC market in dust. After Skylake, Intel planned to launch Kaby Lake and the 10nm Cannonlake processors. The 14nm Kaby Lake were going to launch in 2016 but it seems like we might only see a few mobility chips (mainly the Kaby Lake-Y and Kaby Lake-U series) in Q4 2016 while the remaining parts that include performance desktop and mobility chips will see the light of day in Q1 2017. The Cannonlake processors themselves would arrive in late 2017 for desktops and mobility but it isn’t pointed whether all the chips will make their way to the market in 2017 or we will see a mobile focus push in Q3 ’17 followed by more desktop chips in either Q4 ’17 or early Q1 ’18.
[...]
Just like Skylake, Intel is going to release several SKUs in the Kaby Lake lineup. First up, we have the Kaby Lake-U series which are likely to release in Q3 /Q4 2016 with parts that will include 2+2, 2+2 (Integrated HDCP 2.2). The Kaby Lake-U lineup will consist of chips ranging from 2+2, 2+2 (Integrated HDCP 2.2) and 2+3e (Integrated HDCP 2.2) configurations and will make their way in the market in Q3 2016 for the 2+2 and Q1 2018 for the GT 3e graphics chips. The Kaby Lake-H (4+2) series will arrive in either Q4 2016 or Q1 2017 and will include several SKUs.

BenchLife:

In accordance with the timetable Kaby Lake, we can infer Intel 10nm process of computing will continue next Cannonlake extension, this may mean, to fully see the 10nm manufacturing process Cannonlake platform in the market, the fastest end of the year may be 2018, or It is in early 2019.

I'm a bit unsure how to interpret the Google translated BenchLife quote. Are they talking about a KabyLake-like follow up to Cannonlake when they mention 2018/2019?

 

[ShintaiDK]

Its already posted 2 days ago.

http://forums.anandtech.com/showthread.php?t=2451513

 

[Thanatosis]

That thread has a vague title and is mostly dead. I for one would like to discuss the ramifications of this announcement.


Does this mean TSMC has finally surpassed, or will surpass intel in sram density @10nm? TSMC was claiming 10nm will arrive much earlier than 2H 2017.

 

[ShintaiDK]

TSMC also claimed 10nm would come in the end of 2015 and 16FF in 2013. It happens when it happens. Lets say that.

 

[Thanatosis]

Is samsung going 10nm? They were 14nm back in 2Q 2015, I wonder if they will put up a fight. Samsung's sram density is currently the closest to Intel's but intel still has a significant advantage.

 

[Fjodor2001]

TSMC also claimed 10nm would come in the end of 2015 and 16FF in 2013. It happens when it happens. Lets say that.

What does that roadmap indicate? Start of sampling, mass production, or actual availability in products on the market?
EDIT: I just saw it says "Production", so it must be one of the first two alternatives. Then it's not as far off as if it would have been if it was the last alternative.

Remember that Intel had a similar roadmap before their delays:

 

[Phynaz]

Remember that Intel had a similar roadmap before their delays:

I don't suppose you would want to post that in context?

 

[stingerman]

That thread has a vague title and is mostly dead. I for one would like to discuss the ramifications of this announcement.


Does this mean TSMC has finally surpassed, or will surpass intel in sram density @10nm? TSMC was claiming 10nm will arrive much earlier than 2H 2017.

Intel has kept alive their doped bulk silicon process for some time, but with both 14NM and now 10NM, it's been hard for them to make the kind of advancements they had with their early lead. IBM went the route of of using SOI and more exotic materials for their FinFet entry.

Intel has the challenge of trying to leverage their investment in bulk silicon by using newer doping techniques, where now IBM based process has the advantage of using materials that simplify the production as well provide significantly more interconnects than Intel's and more than TSMC's current 16NM metal interconnect layer. So it will be interesting to see if Apple leverages Samsung and Global Foundries IBM based process to produce significantly larger SOCs at 14NM that will be eventually transferred to 10NM.

The A9X will be very telling if it comes out initially foundered by Samsung.

 

[Systems analyst]

Is samsung going 10nm? They were 14nm back in 2Q 2015, I wonder if they will put up a fight. Samsung's sram density is currently the closest to Intel's but intel still has a significant advantage.

ARM are working with Global Foundries, along with Samsung and TSMC; they are developing 'Physical' IP at 10nm scaling, and also 7nm scaling. This suggests that 10nm is advancing towards production.
Given that ARM finished development of 16nm Physical IP with TSMC one year ago, and that they started working on 10nm with them a year ago, the subsequent success of TSMC 16nm suggests that the parnership is successful; this also suggests to me that Global Foundries and Samsubg will succeed at 10nm.

 

[Arachnotronic]

suggests that the parnership is successful; this also suggests to me that Global Foundries and Samsubg will succeed at 10nm.

Global Foundries couldn't even succeed at 14nm when the process recipe was handed to them. Given that they are going at it alone at 10nm, I think their odds of success are extremely low.

 

[Systems analyst]

Global Foundries couldn't even succeed at 14nm when the process recipe was handed to them. Given that they are going at it alone at 10nm, I think their odds of success are extremely low.

This has been discussed elsewhere in the forum. I think that GloFo bought a design team. Also, they are not alone; ARM are working with them.

 

[imported_ats]

This has been discussed elsewhere in the forum. I think that GloFo bought a design team. Also, they are not alone; ARM are working with them.

Oh yes, because ARM is renowned world over for their process design expertise...

ARM barely plays at all below the synthesis space. They don't really even do P&R or any other physical design. Let alone actual process design.

 

[VirtualLarry]

Oh yes, because ARM is renowned world over for their process design expertise...

ARM barely plays at all below the synthesis space. They don't really even do P&R or any other physical design. Let alone actual process design.

I don't think anyone suggested ARM was doing "process design". Just "physical IP".

Edit: I assumed that the mention of GF acquiring a "design team", was a reference to the personnel they acquired when they bought IBM's fab.

 

[Systems analyst]

Oh yes, because ARM is renowned world over for their process design expertise...

ARM barely plays at all below the synthesis space. They don't really even do P&R or any other physical design. Let alone actual process design.

My understanding is that ARM's standard products, such as Coretex A57, is delivered as RTL. But in the area of physical IP they do more; see for example:
http://www.arm.com/assets/images/PIPD_Logic_IP_Intro_large.jpg
Would you consider this a part of process design, or something at a higher level? Either way, ARM contend that they can provide a lot of stuff to enable a partner to get the best, in some sense, from a particular fab process, so as to minimise cost, or power or maximise performance for the product.

Register files - anyone?
http://www.arm.com/assets/images/rf_large.PNG
This looks like artwork to me; it is optimised for the fab process, thereby improving performance of the end product; however you label this, it is helping the fab company and their customer get the best out of a particular process at that fab. This seems to be pretty low level stuff to me- what do you think?

 

[Fjodor2001]

So here we go for the Intel desktop roadmap:

Skylake: 2015Q3
KabyLake (Skylake + new iGPU): 2016Q4/2017Q1
Cannonlake: 2017H2/2018Q1
KabyLake-like Cannonlake refresh: 2018/2019

Assuming Cannonlake will be 4 cores and minor IPC + frequency increase as usual, this means nothing much will happen on desktop from Intel in the next 4 years.

 

[Phynaz]

So here we go for the Intel desktop roadmap:

Skylake: 2015Q3
KabyLake (Skylake + new iGPU): 2016Q4/2017Q1
Cannonlake: 2017H2/2018Q1
KabyLake-like Cannonlake refresh: 2018/2019

Assuming Cannonlake will be 4 cores and minor IPC + frequency increase as usual, this means nothing much will happen on desktop from Intel in the next 4 years.

Really?

Please post links to Kabylake and Cannonlake reviews showing "not much happening" on desktop.

Anyway, desktop is dead. So dead it's not even worth it to AMD to qualify an Excavator FX CPU.

 

[frozentundra123456]

Well, even though that poster is on my ignore list for good reason, I am afraid he might be right about this one, unless intel decides to bring out a hex core with Cannonlake. They certainly seem to have wrung out nearly all the performance they can (or are willing to) from their quad cores.

 

[Sweepr]

There's a strong chance they will launch 6-8 core Cannonlake parts, of course it's just rumours right now.

KabyLake-like Cannonlake refresh: 2018/2019

Funny trolling, I must admit.
After Cannonlake there's a new CPU architecture (tock).

Assuming Cannonlake will be 4 cores and minor IPC + frequency increase as usual, this means nothing much will happen on desktop from Intel in the next 4 years.

Wait, you're saying they're going to mimic AMD? Vishera crap from 2012 is still being sold today and will be their fastest desktop solution till god knows when in 2016 (assuming no Zen delays). Not even minor IPC bumps for the FX chips.

 

[Fjodor2001]

Really?

Please post links to Kabylake and Cannonlake reviews showing "not much happening" on desktop.

We already know KabyLake will be Skylake + New iGPU. So unless iGPU is important to you, nothing much will happen.

As for Cannonlake, the general expectations seem to be another 5% yearly IPC increase + small frequency bump. None of this is known for sure though. But I guess it's just expectations based on the latest Intel CPU generations. I don't think there is any info available contradicting it either.

Anyway, desktop is dead. So dead it's not even worth it to AMD to qualify an Excavator FX CPU.

I'd not say desktop is dead. In decline though. But I think a lot of that is to blame on the of lack of new applications requiring, and CPUs providing, significantly better performance.

As for AMD Excavator, I agree it's probably not the top focus of AMD, Zen is.

 

[dark zero]

This is Intel monopoly. DEAL WITH IT!

PS: I wanted to post that. Even Apple is still far from Intel

 

[Fjodor2001]

Well, even though that poster is on my ignore list for good reason, I am afraid he might be right about this one, unless intel decides to bring out a hex core with Cannonlake.

So I'm on your ignore list, yet you're reading my posts and agree with what I'm saying. Right...

 

[Fjodor2001]

Funny trolling, I must admit.
After Cannonlake there's a new CPU architecture (tock).

I was referring to the BenchLife article from the OP, saying:

In accordance with the timetable Kaby Lake, we can infer Intel 10nm process of computing will continue next Cannonlake extension, this may mean, to fully see the 10nm manufacturing process Cannonlake platform in the market, the fastest end of the year may be 2018, or It is in early 2019.

I interpret this as that they expect a KabyLake-like follow up to Cannonlake in 2018/2019. But as I wrote before, it's not entirely clear from the Google-translated quote. How would you interpret it?

Wait, you're saying they're going to mimic AMD? Vishera crap from 2012 is still being sold today and will be their fastest desktop solution till god knows when in 2016 (assuming no Zen delays). Not even minor IPC bumps for the FX chips.

No they are following the Intel path that we've seen the last few years. It has nothing to do with AMD.

If anything, AMD will make a significant leap with Zen. Much more so than Intel is making currently going from one CPU generation to the next. But if that will actually materialize is yet to be seen.

 

[frozentundra123456]

There's a strong chance they will launch 6-8 core Cannonlake parts, of course it's just rumours right now.



Funny trolling, I must admit.
After Cannonlake there's a new CPU architecture (tock).



Wait, you're saying they're going to mimic AMD? Vishera crap from 2012 is still being sold today and will be their fastest desktop solution till god knows when in 2016 (assuming no Zen delays). Not even minor IPC bumps for the FX chips.

Well, I will give AMD credit with Zen for at least trying to fight intel head to head on the high end after saying earlier they were not going to try to compete with them. How successful they will be quite obviously remains to be seen, early arriving hype not withstanding, but at least they seem to be giving it a shot. And performance seems to be a priority, in contrast to intel. Again, lets see if they can execute.