Leaked Intel roadmap shows Coffee Lake-R Refresh...

Gideon

Senior member
Nov 27, 2007
475
128
136
#2
A super grainy image very similar to the old roadmap only adding words "refresh"? This looks really dubious until more info confirming this surfaces.

Even videocardz demands for a source - a page that had articles for all the previous Intel roadmap leaks (3 in total in quick succession)
 

exquisitechar

Senior member
Apr 18, 2017
287
209
106
#3
A super grainy image very similar to the old roadmap only adding words "refresh"? This looks really dubious until more info confirming this surfaces.

Even videocardz demands for a source - a page that had articles for all the previous Intel roadmap leaks (3 in total in quick succession)
Indeed. Although, @witeken says that he has a source of his own that backs this up, and that he heard of the refresh before he even saw this slide.
 

jpiniero

Diamond Member
Oct 1, 2010
6,493
295
126
#4
Indeed. Although, @witeken says that he has a source of his own that backs this up, and that he heard of the refresh before he even saw this slide.
Intel did say they were going to make further improvements to 14 nm for Cooper Lake, so the desktop staying on 14++++ makes sense. Possibly get turbo clocks above 5 Ghz.

The mention of Whiskey Lake is a little strange since this seems to be desktop/server lineup, unless they are going to refresh the Pentiums with Whiskey Lake.
 

Gideon

Senior member
Nov 27, 2007
475
128
136
#5
Indeed. Although, @witeken says that he has a source of his own that backs this up, and that he heard of the refresh before he even saw this slide.
Interesting ... I shudder to think what are the implications for Intel if this is true, not just the datacenter, but also the mobile side (particularly Apple), etc ...

This looks extraordinary enough though, to require more evidence.
 
Mar 10, 2004
28,522
238
126
#6
Who cares? If the chips are good, they are good. Doesn't matter if they are 14nm or 10nm.

If Intel can get more out of the 14nm node, great.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,385
166
126
#7
I think the biggest thing is Elkhart Lake.
- No Gemini Lake Plus
- No Jasper Lake
- No Mercury Lake
- No Meteor Lake

If the roadmap is true, it is probably indicating Intel skipping 10nm for 7nm.

Intel's 10nm started in 2017. With HVM production starting on top of 7nm start in 2H2019. With the better tools from Nikon, Intel can probably produce 7nm after a year from start.
 

arandomguy

Senior member
Sep 3, 2013
470
16
116
#8
What I'm curious about all this the decision making resulting in the current uarch stagnation, both the CPU side and GPU side, since Skylake in 2014. They've had to design and do new tape outs regardless for all subsequent Skylake iterations. At least Cannonlake generation is completely with Icelake being said to have been finalized as well.

Also if they do hypothetically skip to 7nm at this point does that mean Icelake (at least for mainstream) and subsequent iteration get skipped as well? Interestingly this roadmap does not name Icelake.
 

exquisitechar

Senior member
Apr 18, 2017
287
209
106
#9
Who cares? If the chips are good, they are good. Doesn't matter if they are 14nm or 10nm.

If Intel can get more out of the 14nm node, great.
I would somewhat agree if we didn't just keep getting Skylake on further and further improved 14nm. 10nm's efficiency and density would still be sorely missed.
I think the biggest thing is Elkhart Lake.
- No Gemini Lake Plus
- No Jasper Lake
- No Mercury Lake
- No Meteor Lake

If the roadmap is true, it is probably indicating Intel skipping 10nm for 7nm.

Intel's 10nm started in 2017. With HVM production starting on top of 7nm start in 2H2019. With the better tools from Nikon, Intel can probably produce 7nm after a year from start.
Interesting, some others have suggested the same thing.
 

french toast

Senior member
Feb 22, 2017
943
59
106
#10
That would be quite the comeback if true, 7nm intel 2H 2020??..seems a bit tight to be possible.
What would AMD be rocking..7nm + EUV....meaning intel would presumably take back the lead by end of 2020 (process).

10nm is dead already, 10nm++ wouldn't even be 10nm anymore?
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,385
166
126
#11
That would be quite the comeback if true, 7nm intel 2H 2020??..seems a bit tight to be possible.
https://www.semiwiki.com/forum/content/7544-7nm-5nm-3nm-logic-current-projected-processes.html
https://www.semiwiki.com/forum/attachments/content/attachments/21825d1529099842-slide5.jpg

Intel's 7nm from Scotten Jones is on par with Samsung's 3nm GAAE/GAAP. Intel might hop back onto the more aggressive shrink style of 14nm to 10nm. So the numbers from Jones are the worst case for Intel.
 

Dayman1225

Senior member
Aug 14, 2017
901
90
96
#12





Pick your poison because on of these must be right, right?
 

jpiniero

Diamond Member
Oct 1, 2010
6,493
295
126
#13
Also if they do hypothetically skip to 7nm at this point does that mean Icelake (at least for mainstream) and subsequent iteration get skipped as well? Interestingly this roadmap does not name Icelake.
I don't think they are skipping 10 nm, they can't because their 7 nm products almost have to be dependent on 10 nm being available. 2H 2020 is possible for 7 nm but rather unlikely. It's more 2021 I think.

Not surprised they would cancel Icelake Server; I figured it was going to be a paper launch at best. Perhaps they got some new Intel (heh) which led them to believe it wasn't going to work in confusing the market against Rome anyway.
 

exquisitechar

Senior member
Apr 18, 2017
287
209
106
#14
Pick your poison because on of these must be right, right?
Heh, nice. I'd probably consider one of these over the one in the OP. It has some glaring issues...still thought it was worth posting, though.
 

mikk

Platinum Member
May 15, 2012
2,430
48
126
#15
This is most likely a selfmade Roadmap from someone.
 

TheGiant

Senior member
Jun 12, 2017
491
126
86
#16
Personally I couldn't care less if its true if
  • it brings more frequency (which I consider hard but maybe 5,2GHz)
  • it brings moar coarz (from 8 to 10 or even 12 on ring bus so gaming doesn't suffer)
  • decreases power consumption with same amount of coarz
  • it brings moar IPC

I don't get this nm issue. We forgot what is the goal and the tool. The world issue.
 
Oct 14, 2003
6,209
324
126
#17
Intel's 10nm started in 2017. With HVM production starting on top of 7nm start in 2H2019. With the better tools from Nikon, Intel can probably produce 7nm after a year from start.
:rolleyes:

No fabs have skipped a process, because next generation process always build upon the previous generation. It's like saying let's skip grade 8-12 since I failed, and head straight to university.

Many people are pointing to Intel's needless focus on density for 10nm delays. Do you think a further focus in density at 7nm will make it easier?

No, if the roadmap is true then we won't see 10nm until 2020, its that simple.

I would somewhat agree if we didn't just keep getting Skylake on further and further improved 14nm.
This is because somewhere, someone important at Intel was too arrogant and refused to believe 10nm will cause such an issue for them. If you are absolutely confident that the next gen process will succeed, then it'll be a waste to make a backup based on 14nm. That's why 14nm refresh exist, rather than an updated architecture on 14nm.

This isn't about tech, and what's possible or not, but people running the company and the culture that makes it.

At this rate, the next fab player to fail after Global Foundries is Intel. So we'll have Samsung and TSMC left. Seeing how things are going, in just a few years we might just have TSMC as the only one making latest process(until they stop too of course).

I don't get this nm issue. We forgot what is the goal and the tool. The world issue.
The gains might be smaller than ever, and the efforts required greater on top of that, but a new process is still the best and surefire way to get improvements on microprocessors.

Think of an another possibility. Intel did not have a crazy focus on density and they got 10nm out in time. Competitive pressure from AMD and others would have caused them to create better 10nm products. It would be based on new IP. It would be more energy efficient, better performing, lower cost and have newer tech than 14nm could have ever brought.
 
Last edited:

NostaSeronx

Platinum Member
Sep 18, 2011
2,385
166
126
#18
No fabs have skipped a process, because next generation process always build upon the previous generation.

Many people are pointing to Intel's needless focus on density for 10nm delays. Do you think a further focus in density at 7nm will make it easier?
GlobalFoundries has skipped processes. The next generation process is not always built upon the previous generation.

10nm is delayed regardless of density or not. Intel doesn't have the tools to support HVM 193i FEOL/MOL on 10nm. While by the time for 7nm they will till at least their 5nm node. So, 7nm gets the focus and launches first while 10nm gets custom foundry treatment.

Intel like Samsung doesn't need all their products on the nodes they provide.

Intel implements 1st Gen post-FinFET on 7nm then 2nd Gen post-FinFET on 5nm. Restarting the two year cycle. By the 6 month mark for 5nm, Canon Nanotechnologies 6-cluster JFIL for 1nm nodes and up should be ready.
 
Oct 14, 2003
6,209
324
126
#19
Intel like Samsung doesn't need all their products on the nodes they provide.
Intel's Custom Foundry is a joke, and an excuse. That's another thing they are doing wrong.

They should have focused on their bread-and-butter PC business and expanded from there. Their process problems started all the way back in 22nm when it started getting delays and we started to get minimal improvements on the PC side. Then 14nm happened and got much worse.
 

jpiniero

Diamond Member
Oct 1, 2010
6,493
295
126
#20
:rolleyes:
Many people are pointing to Intel's needless focus on density for 10nm delays. Do you think a further focus in density at 7nm will make it easier?
It's possible they have finally come to the conclusion that it is not really fixable without EUV, and by the time EUV is ready the 7 nm designs should be too, and they have focused their efforts on completing that, and perhaps porting the dependent products to EUV rather than spin their wheels trying to fix Icelake.
 
Oct 14, 2003
6,209
324
126
#21
It's possible they have finally come to the conclusion that it is not really fixable without EUV, and by the time EUV is ready the 7 nm designs should be too, and they have focused their efforts on completing that, and perhaps porting the dependent products to EUV rather than spin their wheels trying to fix Icelake.
In the extremely rare case, where the company is working on a process that's far behind the rest of the industry it can be "skipped", just because the work has been done by others already. Even in that case its a lot of risk as new processes cost a lot upfront and you are hoping moving your products to the newer process will pay itself off.

None of that is true in the case of Intel.

Remember the reason they introduced Tick/Tock methodology of development? Moving to a new architecture on the same process brings uncertainties to development. Moving to a new process on the same architecture is also a risk. By doing only one at a time, you are on a stable foundation for at least one of them.

Not only 7nm would have zero track record(at least 10nm has the broken Cannonlake chips for production), having a new architecture adds more risk. 7nm also needs to bring noticeable power reduction and performance improvements, which aren't guaranteed by using EUV.
 

Nothingness

Golden Member
Jul 3, 2013
1,905
35
106
#22
Personally I couldn't care less if its true if
  • it brings more frequency (which I consider hard but maybe 5,2GHz)
  • it brings moar coarz (from 8 to 10 or even 12 on ring bus so gaming doesn't suffer)
  • decreases power consumption with same amount of coarz
  • it brings moar IPC

I don't get this nm issue. We forgot what is the goal and the tool. The world issue.
The problem is that Intel co-designs their new CPU and process. So improvements are attached to a process and if a process is a failure making the new CPU will be an issue. Back porting to an older process is not always an option and requires a lot of work. And last thing, when you have started converting fabs to a new process that doesn't deliver, you will have output issues, and rumors have it that Intel has capacity issues; this means higher prices for you.

That being said in your list this mostly affects IPC: significant IPC increases comes with a new micro architecture which Intel had scheduled for 10nm.
 

jpiniero

Diamond Member
Oct 1, 2010
6,493
295
126
#23
There's also EMIB and possibly Feveros in the mix too, that would be quite a lot all at once.

If Intel does end up shipping a decent amount of Icelake U/Y at the end of 2019 (even if the majority is Comet Lake and Whiskey Lake/Refresh), perhaps they could release products in 2020 using EMIB with the 10 nm tiles kept small.
 


ASK THE COMMUNITY

TRENDING THREADS