[Anandtech] Intel's Architecture Day 2018

Page 14 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Intel's Architecture Day 2018: The Future of Core, Intel GPUs, 10nm, and Hybrid x86

I know some bits of this were getting some mention in other threads, but it seems like it was big enough to warrant it's own discussion.

Highlights for me were a new architecture (finally) Sunny cove, that looks to have potential (4-5 wide allocation) to deliver Intels first real IPC improvement in MANY years, and also what should be a nice improvement in the the IGP, and a look at Intels multichip solutions(Foveros) and Intel Big-little.

The Q&A was also somewhat interesting, like where they admit being node-locked on their designs really messed them up, and they will be more node agnostic in the future.

Overall, more actual news than we have seen from Intel in some time.
 
  • Like
Reactions: Burpo
Mar 10, 2006
11,715
2,012
126
Comet is very real but also 14 nm based.



I have serious doubts about that. I think you're more likely to see them try to rush out 7 nm instead once (if?) possible. Or using a 14 nm Willow Cove based CPU cores with a product using Forveros (and not monolithic)

Cool, you know more about Intel's product roadmap than Raja Koduri? Please, tell us more about it!
 
  • Like
Reactions: Dayman1225 and mikk

Spartak

Senior member
Jul 4, 2015
353
266
136
Highly annoying how jpiniero is trying to make his own speculation come off as fact. Even if it's based on some info/leaked ES, Intel's roadmap has likely never been as unstable and unpredictable as the coming year.

There's a good chance they still have a plan A and plan B as we speak. But since we got a high profile presentation and very positive rumblings about their 10nm 'alternative' i'm pretty bullish their plan A is close to secured.

And plan A isn't a 10-core skylake on 14nm to be clear :)
 
Last edited:

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Highly annoying how jpiniero is trying to make his own speculation come off as fact. Even if it's based on some info/leaked es Intel's roadmap has likely never been as unstable and unpredictable as the coming year.

There's a good chance they still have a plan A and plan B as we speak.
Haha yeah. I’m of the opinion some sort of 14nm desktop stop gap will come this year, but I need more convincing than a forum post saying CML 10c is coming for me to believe 10c CML is real.

Also I don’t think Sunny Cove will ever make it to desktop and they will wait for Willow Cove to bring 10nm to DT.

And I agree with your point about contingency plans, they seem to have learned from 10nm mistakes, albeit a little late.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Cool, you know more about Intel's product roadmap than Raja Koduri? Please, tell us more about it!

Basically I think he's lying because if he said anything otherwise it would get people going. But I guess you shouldn't rule out Intel doing 10 nm EUV even though BK said they weren't doing it. Maybe they will now more as a fallback in case 7 EUV was too much.

I actually think Icelake is going to be Intel's last monolithic product anyway.

Why do you seem so convinced that 10nm will never work out for desktop...?

Because to do that they would have to:

A.) Actually fix the yield
B.) Get the clock speeds high enough to make it worth it.

I am unconvinced they can do A, and with the core design unhooked from the process they could in theory release Willow Cove on 14 nm and be faster with higher clocks albeit very toasty.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Because to do that they would have to:

A.) Actually fix the yield
B.) Get the clock speeds high enough to make it worth it.

I am unconvinced they can do A, and with the core design unhooked from the process they could in theory release Willow Cove on 14 nm and be faster with higher clocks albeit very toasty.
Yield seems to improving considering they’ve committed to launching Icelake Mobile 4/8c GT2 64EUs later this year, Lakefield, Nervana Inference ASIC and Snow Ridge 5G Base station SoC also later this year, and based on Architecture day and their CES presentation they seem a lot more confident about 10nm. Last year they at CES they didn’t talk about 10nm at all and released a GPU-less Cannonlake, this year CES had a lot of 10nm.

It’s quite clear Charlie was wrong about 10nm being cancelled IMO.

Also it’s VERY unlikely intel insert EUV into 10nm.

All we can do is wait and see.
 

Adawy

Member
Sep 9, 2017
79
24
41
Basically I think he's lying because if he said anything otherwise it would get people going. But I guess you shouldn't rule out Intel doing 10 nm EUV even though BK said they weren't doing it. Maybe they will now more as a fallback in case 7 EUV was too much.

I actually think Icelake is going to be Intel's last monolithic product anyway.



Because to do that they would have to:

A.) Actually fix the yield
B.) Get the clock speeds high enough to make it worth it.

I am unconvinced they can do A, and with the core design unhooked from the process they could in theory release Willow Cove on 14 nm and be faster with higher clocks albeit very toasty.

https://fuse.wikichip.org/news/2004/iedm-2018-intels-10nm-standard-cell-library-and-power-delivery/

Did you read that article?
 
  • Like
Reactions: Arachnotronic

Spartak

Senior member
Jul 4, 2015
353
266
136
He also continues to igore Sunny Cove can either be backported to 14nm, we could see some form of half node if 10nm failed, or they got 10nm+ working in time at acceptable yields.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Haha yeah. I’m of the opinion some sort of 14nm desktop stop gap will come this year, but I need more convincing than a forum post saying CML 10c is coming for me to believe 10c CML is real.

Also I don’t think Sunny Cove will ever make it to desktop and they will wait for Willow Cove to bring 10nm to DT.

And I agree with your point about contingency plans, they seem to have learned from 10nm mistakes, albeit a little late.

The difference with the previous three years is they cannot afford anymore to not have contingency plan in place this time around.

And let's not kid ourselves, 8 cores was never intended for desktop 14nm either. six core maybe not even.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
The difference with the previous three years is they cannot afford anymore to not have contingency plan in place this time around.

And let's not kid ourselves, 8 cores was never intended for desktop 14nm either. six core maybe not even.
Six core maybe, 8c def not.
 

exquisitechar

Senior member
Apr 18, 2017
655
862
136
IMO, there won't be a backported 14nm Sunny Cove in 2019. I just don't interpret what was said during Architecture Day to mean that. I could be wrong, who knows anything for sure with how quiet Intel is being about desktop? I think that we will get proper 10nm (or 10nm+) DT in 2020, and it may be Willow Cove based, skipping Sunny Cove for DT. To me, it is clear they are making progress on 10nm, there will be no half node or cancellation business going on - it will just take time.
 

jpiniero

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

Yep. But you can design all you want and it doesn't matter if you can't really produce that much because the yield sucks.

He also continues to igore Sunny Cove can either be backported to 14nm, we could see some form of half node if 10nm failed, or they got 10nm+ working in time at acceptable yields.

I think Sunny Cove is the last of the non-portable designs. You could see WIllow Cove on 14 nm in theory however.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Yep. But you can design all you want and it doesn't matter if you can't really produce that much because the yield sucks.

I think Sunny Cove is the last of the non-portable designs. You could see WIllow Cove on 14 nm in theory however.

This is complete nonsense. There is no such thing as non-portable designs. You think Keller & co are making things up?

If this is the level of your reasoning you might as well stop contributing, as it's just fluff.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
This is complete nonsense. There is no such thing as non-portable designs. You think Keller & co are making things up?

Umm, Intel made it a big point that previous designs were tightly coupled with a process. Which probably also includes Sunny Cove as the last of the non-portable designs.
 

exquisitechar

Senior member
Apr 18, 2017
655
862
136
Intel must have competitive 10nm chips on the desktop in 2019, imo.

2020 is too late.
Why "must"? They'll survive and even thrive, overall, even if they get pummeled in desktop for 6+ months...and come back with a vengeance, looking at 10nm's characteristics and Sunny Cove's extensive changes.
Umm, Intel made it a big point that previous designs were tightly coupled with a process. Which probably also includes Sunny Cove as the last of the non-portable designs.
Rather than "non-portable", it would probably be more accurate to say that it would be somewhat difficult to port those designs, but not impossible.

And yeah, I have to agree with you that Sunny Cove is that kind of design. It seems to me that Intel acknowledged that tightly coupling microarchitectures with processes is what got them into this mess - Sunny Cove, specifically, being the "non-portable" kind, hence why we've had years of Skylake. It's future microarchitectures that will not be process dependent. Not that I'm saying that there's no chance that Sunny will be backported, maybe Intel had the foresight to make the decision to do so a while ago, despite the fact that the idea was shot down at first. I think it was Francois Piednoel that said that he advocated for it (backporting Icelake) but that it didn't work out.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Umm, Intel made it a big point that previous designs were tightly coupled with a process. Which probably also includes Sunny Cove as the last of the non-portable designs.

For the zillionth time Intel has stated this was not because of the design itself or the process but the abstraction tools lacking. These tools are now in place and can be used for any architecture. This is not me stating this, this is literally stated very specifically by intel.

Go read the interview again since it seems the crucial parts flew right over you.

Whether there is a sufficient time window for sunny cove on 14nm this year is an entirely different matter which depends on when they had the tooling in place and the lead time to manufacturing. But suggesting they'd backport Willow Cove to 14nm next year is borderline trolling.
 
Last edited:

Spartak

Senior member
Jul 4, 2015
353
266
136
Rather than "non-portable", it would probably be more accurate to say that it would be somewhat difficult to port those designs, but not impossible.

And yeah, I have to agree with you that Sunny Cove is that kind of design. It seems to me that Intel acknowledged that tightly coupling microarchitectures with processes is what got them into this mess - Sunny Cove, specifically, being the "non-portable" kind, hence why we've had years of Skylake. It's future microarchitectures that will not be process dependent. Not t the fish hat I'm saying that there's no chance that Sunny will be backported, maybe Intel had the foresight to make the decision to do so a while ago, despite the fact that the idea was shot down at first. I think it was Francois Piednoel that said that he advocated for it (backporting Icelake) but that it didn't work out.

Has seriously nobody read and understood the interview?

Intel has stated they couldnt fry the fish not because the wood was wet or there was something wrong with the fish, but they didnt have matches.

You guys keep claiming you can't fry this particular fish. It's absurd.
 
Last edited:

exquisitechar

Senior member
Apr 18, 2017
655
862
136
You guys keep claiming you can't fry this particular fish. It's absurd.
I know they can, I said as much. There being a sufficient time window is what I'm not sure of.
For the zillionth time Intel has stated this was not because of the design itself or the process but the abstraction tools lacking. These tools are now in place and can be used for any architecture. This is not me stating this, this is literally stated very specifically by intel.
Right, that completely makes sense from what I know - although the coverage of Architecture Day that I read didn't go in-depth concerning this. What I'm not convinced of is that they had the tools in time to backport Sunny, but I am much more open to the idea than I was before after looking into it further.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
That I do agree on. We don't know if there was/is sufficient time to backport Sunny Cove now that they can.

But even if it means a delay to 20Q1 that would be much more preferable than a stopgap 10-core skylake in 19Q4.