Discussion Intel current and future Lakes & Rapids thread

Page 74 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
It aint a CPU lineup - its a platform. All I can say without giving away the whole thing.

Charlie did say earlier that he thought that Cooper and Icelake Server would be on different platforms. Not really sure how that would work but if Intel will release Cooper Lake products later that ran on Whitley I guess that would make sense?
 

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
2 cores with 1mb L2 and 8mb L3?

Looks like 128mb of edram also.

Actually it reports 1 core with 2 threads which seems wrong.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
2 cores with 1mb L2 and 8mb L3?

Probably a quad core die with only two enabled. Icelake has the L2 increased to 512 kb/core.

There is no guarantee of course that Intel will actually ship a model with 64 EU enabled but it in theory exists.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
The first fully enabled GT2 in the wild, nice. Yes they could release the first wave of ICL-U with 48 EUs because of poor yields, who knows.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
The first fully enabled GT2 in the wild, nice. Yes they could release the first wave of ICL-U with 48 EUs because of poor yields, who knows.

Plausible. But 4C+64EU, you're going to be pushing 100 mm2 now on a node that doesn't work. Aggressive yielding is likely going to be needed regardless of what they end up settling on.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Plausible is also a different market segmentation between the different SKUs.

i7= 64 EUs
i5= 48 EUs
i3= 32 EUs

Something like this.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Why is the report sketchy?

If the Bluefin Research group has went and talked to PC suppliers about it, then its quite credible. More often than not, rumors and leaks have been correct.

Isn't that still 2Q 2019?

I doubt we'll see it that early. Amberlake systems are just starting to be announced now, and we're not seeing Whiskey Lake yet. Probably Computex launch, system announcements in August/September, and being on shelves 1-2 months after that.
 
Mar 10, 2006
11,715
2,012
126
Why is the report sketchy?

If the Bluefin Research group has went and talked to PC suppliers about it, then its quite credible. More often than not, rumors and leaks have been correct.



I doubt we'll see it that early. Amberlake systems are just starting to be announced now, and we're not seeing Whiskey Lake yet. Probably Computex launch, system announcements in August/September, and being on shelves 1-2 months after that.

BlueFin is one of the best channel check services around.
 
  • Like
Reactions: DooKey

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Bloomberg has more detail from the report.
smNNAxa.png
 
  • Like
Reactions: DooKey

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
^^^

Much better. So we'll see, Intel obviously has some incentive to get going on 10 nm with the 14 nm capacity issues and the monstrosity that is Cooper Lake looming.

And while you can still presume that the first product to HVM would be Icelake U/Y, there's no guarantee. It could be the FPGAs.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
^^^

Much better. So we'll see, Intel obviously has some incentive to get going on 10 nm with the 14 nm capacity issues and the monstrosity that is Cooper Lake looming.

And while you can still presume that the first product to HVM would be Icelake U/Y, there's no guarantee. It could be the FPGAs.
How big is Cooper Lake supposed to be or how many cores?
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
How big is Cooper Lake supposed to be or how many cores?

It's basically two Cascade Lake-SP dies fused together with EMIB, or at least that's the expectation. So the top would be 56.

One Skylake-SP XCC die is 694 mm2.
 
  • Like
Reactions: Ajay

JDG1980

Golden Member
Jul 18, 2013
1,663
570
136

This proves that most of the "analysts" making stock decisions don't know a damn thing about technology.

If Intel does ramp up 10nm in April 2019, it's not going to start with large monolithic server chips. That's a nonstarter. Intel hasn't even released Cascade Lake yet, let alone Cooper Lake, and those are both 14nm+++ products. We might see some low-power Icelake products in mid-2019 if we're lucky; Intel already said the desktop versions wouldn't be available until sometime in 2020. Or we might just see more Cannonlake parts.

The bull run on AMD is largely based on their projected increase in market share of high-margin server chips (Epyc). The fact remains that AMD is going to have TSMC-fabbed 7nm server chips in mid-2019, while Intel won't have 10nm server chips until sometime in 2020. This "research report" doesn't change that, and therefore it makes little sense for them to have affected the stock price. This simply demonstrates the irrationality of the stock market.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
4-6 months from production to store shelves is too long, usually 3 months are enough. In some cases 2 months.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
4-6 months from production to store shelves is too long, usually 3 months are enough. In some cases 2 months.

3 months to produce, but then it takes another month or two for the OEMs to receive the chips, get the laptops assembled, and shipped to retailers.
 
  • Like
Reactions: lightmanek

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
3 months to produce, but then it takes another month or two for the OEMs to receive the chips, get the laptops assembled, and shipped to retailers.


2 months from production till launch, for notebooks 1 additional month until the first devices becomes available.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
The bull run on AMD is largely based on their projected increase in market share of high-margin server chips (Epyc).

It's not entirely due to server. The AMD bull run depended on AMD being completely positive without *any* downsides. Now presumably it won't be as dominant, so it'll temper its expectations down a bit. The original expectation wasn't realistic of course.

Look at the valuation comparison between Nvidia and Intel. The latter has profit and net revenue at more than 5x the former, yet there's only 1.2x difference in valuation. It's same with Netflix, and Facebook. The darling gets all the attention, and even 2nd place gets a fraction of it.

You can see the same with athletes. The first place earns like he owns the world and sometimes the 2nd seems almost like he has a regular career.

Icelake laptops on store shelves until September-October

Looking at Amberlake and Whiskey Lake suggests this looks about the time when we'll see Icelake.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
By the way I do think what Intel ended up doing was more or less cutting the quad patterning layers back down to double, and regressing the density as needed. Sounded like that actually worked.

My guess is that it will be around 100 mm2 for the 4C+64EU U/Y model, assuming that is the first out the gate and accounting for the density hit.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
By the way I do think what Intel ended up doing was more or less cutting the quad patterning layers back down to double, and regressing the density as needed. Sounded like that actually worked.

Well, they were saying 5-6 patterning not just quad.

My guess is that it will be around 100 mm2 for the 4C+64EU U/Y model, assuming that is the first out the gate and accounting for the density hit.

Their main chips NEVER got the density benefit, it was always for the little cores like Atom and others, maybe FPGA?

The Core line always had slightly less than 50% reduction with new process, so little over half the size. When they claimed density lead with 14nm, it was always the maximum attainable. Like with 14nm Airmont, it got an amazing 64% reduction in size compared to Silvermont. Unlike Broadwell, which I'm getting greater than 55% core size(so only <45% reduction) compared to Haswell at 22nm.

I assume for the main core line, the rejigged 10nm didn't affect the density at all.
 
Last edited: