Discussion Intel current and future Lakes & Rapids thread

Page 154 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
Hmm, makes me wonder how viable EMIB will ever be if Intel continues having such problems. AMD doesn't even seem to try.

EMIB will work fine for connecting desktop and server chips in MCM packaging, and maybe 45/55W H mobile chips. When you want to save power in a milliwatt level as with ultrabooks and needing modern standby, the whole picture changes.

For what its worth, the architecture day roadmap had Sunny/Willow/Golden coming every year.
 
Last edited:
  • Like
Reactions: ikjadoon

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,653
146
I doubt AMD is able to reach Golden Cove IPC in 2021. Intel currently has a 18% IPC lead over AMD with Sunny Cove. Next year it will probably increase to about 25% with Willow Cove. AMD will decrease the deficit with Zen 3 but they surely won't close down the gap. Golden Cove should bring another huge leap over Willow Cove. AMD would need two IPC steps with Zen4 in one generation to close the gap. Some people seem to forget that Intel has a manufacturing problem and not a design problem.

Not fully close down the gap no, but it'll make huge strides to doing so.

Also, 'huge' is relative. Golden Cove is big, but I wouldn't call it huge.

Lets say that if everything went as planned for all the Zen archs - specifically the ones already released - Golden Cove would look pretty middle-ish. But some things got pushed up and by current standards Golden now looks on the higher end of IPC bumps.
 

coercitiv

Diamond Member
Jan 24, 2014
7,225
16,982
136
I'm a little surprised by this response. It sounds like you are saying canning 10nm would have been a better idea.
I was quite surprised myself too, as until yesterday I was quite convinced that keeping 10nm was a good idea even if all it did was buying PR points.

But I never imagined Intel would not be able to satisfy 14nm demand in Q4 2019. That was quite the shock for me, and I already mentioned I take it as a lesson, not as a critique to Intel.

alternative would mean waiting for 7nm, which never will happen because that's a 5x density jump from 14nm and that nearly guarantees complete end.
I agree here, in the sense that management may have chosen to keep 10nm and eventually make it work as insurance.

And secondly, sticking with Cometlake and Rocketlake instead of Icelake, and Tigerlake which is far superior.
Tigerlake maybe, Icelake is a draw in the CPU department. You gain IPC, you lose clocks, and AFAIK efficiency wise it's a draw as well.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
Tigerlake maybe, Icelake is a draw in the CPU department. You gain IPC, you lose clocks, and AFAIK efficiency wise it's a draw as well.

You can't ignore the graphics either. Icelake would have ended up with the CPU portion being 2x as large and GPU 3x as large. So they'll end up with a 25% larger chip(150 vs 120mm2) with half the graphics or same graphics and 2x large die(240mm2).

Don't forget about other 10nm chips coming either. Lakefield is going to be another big one, but there's Snow Ridge/Icelake-SP and Tremont derivatives.

The ideal part would have been the previous management being smart and not overdoing on 10nm so they couldn't possibly meet the goals. But they have to deal with what they have.

Little about volume: https://liliputing.com/2019/11/one-netbook-one-gx-handheld-gaming-pc-coming-next-summer.html

Tigerlake coming few months earlier. Depending on the sources they talk about 1H of 2020 or Summer.

Representatives of One Netbook during an interview with PCWatch has said Icelake-Y is skipped because the -Y chip is reserved only for Apple and Microsoft, not even big ones like Lenovo. One new state of the art fab, Fab 42 is coming online in 2020 for 10nm/7nm(and an existing one is moving to 14nm also in 2020). Will be needed for Snow Ridge/Icelake-SP and Tremont. So slowly but surely it'll improve, but will continue to be stormy.
 
  • Like
Reactions: coercitiv

DrMrLordX

Lifer
Apr 27, 2000
22,696
12,650
136
...uh, is Golden Cove for Alder Lake NOT confirmed?

I could have sworn it was...

I see a lot of commentary about Alder Lake being Golden Cove, but I've never assumed that would actually be the case. I figured the first Golden Cove would be Meteor Lake in 2022 on 7nm.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
Guesses on Tigerlake if they move to the split L2 configuration.


1575117658097.png
(I expect the current L2 cache to turn into L2-D)

So L2 cache used to be "uncore", but starting with Nehalem it became part of the core. Based on the shape of the L3 cache for Icelake, even L3 is going in that direction. Actually, it might have started with Skylake.

It makes sense, because some form of an L4 cache, or super fast system memory level is coming, but very slowly. eDRAM on-die has serious limits on scaling, and new memory technologies(STT-MRAM/2T SRAM) are not yet viable. eDRAM off-die raises costs and its not dense enough for system memory. HBM is more dense(but slower than eDRAM), but still has the problem of being difficult to implement and high cost.

But we'll see premium client and server use HBM/eDRAM just because that's the way to go for higher performance. Intel will likely ditch eDRAM and move to HBM for future products.

1575117704628.png
(The above "simple" changes took extra year to implement. Why do I call it "simple"? Because just below the extended AVX has regular AVX. Just to the left of the extra L2 cache it has the regular L2 cache. To the right of extra AVX is where the L3 resides. It looks like a simple copy paste job! If they put in extra work, the AVX-512 and L2 cache could have been seamlessly integrated and faster, but that's something that takes 3+ years.)

Making a new architecture is *hard*. It cannot be emphasized enough. Though with likely better tools and methodologies more custom work is being done. In the Northwood days, doubling L2 cache was a change that came occasionally. Nowadays, we see new uarchs happening every 1-2 years!
 

ksec

Senior member
Mar 5, 2010
420
117
116
What is the latest Roadmap for Intel 's Next Gen 35/45W Mobile CPU?

Is it still 14nm, or are there any 10nm? And When is it coming?
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,653
146
I see a lot of commentary about Alder Lake being Golden Cove, but I've never assumed that would actually be the case. I figured the first Golden Cove would be Meteor Lake in 2022 on 7nm.
I'm pretty sure Intel specified Golden Cove is coming in 2021. They didn't say what with though now that I think about it.
 

jpiniero

Lifer
Oct 1, 2010
16,493
6,983
136
What is the latest Roadmap for Intel 's Next Gen 35/45W Mobile CPU?

Is it still 14nm, or are there any 10nm? And When is it coming?

Intel appears to be transitioning the U line to cover the 35/45W mobile market. There might be actual dedicated 14 nm Rocket Lake U models advertised with that TDP but they don't necessarily have to do that since the OEM could just raise the PL1 to 35 W. Pretty sure there's at least one 6 core Comet Lake U laptop available that is doing just that.
 

uzzi38

Platinum Member
Oct 16, 2019
2,746
6,653
146
Also, @DrMrLordX you said something about AMD not trying to create something like EMIB - they won't even bother.

Not when they can just use SOIC
 

ksec

Senior member
Mar 5, 2010
420
117
116
Intel appears to be transitioning the U line to cover the 35/45W mobile market. There might be actual dedicated 14 nm Rocket Lake U models advertised with that TDP but they don't necessarily have to do that since the OEM could just raise the PL1 to 35 W. Pretty sure there's at least one 6 core Comet Lake U laptop available that is doing just that.

Which is to say hopeless for 10nm?

Rather unfortunate being stuck with Intel on Apple ecosystem.
 

Vattila

Senior member
Oct 22, 2004
817
1,450
136
Again, this is not about whether 10nm is functional or not, it's about whether or not Intel's 10nm capacity & yields are able to offset the huge 14nm demand they're seeing, so huge they are publicly apologizing, so huge that Dell is forced to announce they're missing financial targets. […] But I never imagined Intel would not be able to satisfy 14nm demand in Q4 2019. That was quite the shock for me, and I already mentioned I take it as a lesson, not as a critique to Intel.

Slightly OT, but I think it is worth commenting on this. "Demand" will increasingly be used by Intel as a smokescreen for their manufacturing issues.

The real picture is that Intel manufacturing is unable to supply a declining PC market, partly because of major manufacturing problems (the 10nm debacle, in particular), huge commitments to Apple on modem production, as well as the relentless progress by AMD towards more cores and more performance, increasing the required wafer capacity to produce the same number of chips. This competitive pressure has led Intel to overextend themselves, resulting in failure to fulfil commitments, and subsequently, in the public apology and filing of the 8-K report.

Remember, demand is not the same as growth. Obviously, there is "demand" when customers get on the phone to Intel demanding something is done to fix shipment delays. But this does not mean that the purchase orders are any bigger, though. As broadly reported, and admitted in their public letter, Intel has problems fulfilling existing purchase plans and commitments, causing delays.

Regarding claimed strength in the PC market, it is basically a slower decline than previously forecasted. IDC is now projecting a flat 2019 (+0.5%), which is slightly better than earlier projections. There is some effect from Windows 7 end-of-life, and there are some segments that are growing, such as 2-in-1 and slim notebooks, content creation and gaming. But overall, the PC market is still forecasted to shrink 2.6% per year in the period 2019-2023. Much of that decline is attributed to slower tablet sales, though.

"On the heels of strong growth during the third quarter of 2019 (3Q19) and changing dynamics in the market for personal computing devices (PCDs), International Data Corporation (IDC) has raised its outlook for the remainder of 2019. According to a new forecast from the Worldwide Quarterly Personal Computing Device Tracker, overall PCD shipments will reach 407.7 million by the end of 2019, up 0.5% from 2018. However, despite this recent growth, long-term forecasts remain negative as the market will dip to 366.7 million units in 2023 with a compound annual growth rate (CAGR) of -2.6% from 2019–2023. […] Though other form factors such as traditional desktops and notebooks will continue to decline, IDC anticipates a growing share of those form factors will cater to emerging demand for creator PCs as well as sustained demand for gaming PCs."

idc.com

getfile.dyn


Personal Computing Device Forecast, 2019-2023 (shipments in millions)
Product
2019 Shipments*
2019 Share*
2023 Shipments*
2023 Share*
2019-2023 CAGR
2-in-1
50.5
12.4%
72.4
19.7%
9.4%
Desktop + Desktop Workstation
93.8
23.0%
76.5
20.9%
-5.0%
Notebook + Mobile Workstation
74.9
18.4%
50.3
13.7%
-9.5%
Slate Tablet
107.0
26.3%
67.3
18.3%
-11.0%
Ultraslim
81.4
20.0%
100.3
27.3%
5.3%
Total
407.7
100.0%
366.7
100.0%
-2.6%
 
  • Like
Reactions: guachi and yeshua

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
Intel appears to be transitioning the U line to cover the 35/45W mobile market.

I wonder if the leak just doesn't show the -H models, or they are really sticking with Comet Lake for another year.

Even if you increase TDP, the -H series needs more cores than 6 and Rocketlake doesn't deliver that. Maybe that's just undecided and they'll just make the -H series by reusing the -S chips.

On a separate note, Intel says EMIB powered CPUs and FPGAs are used in nearly 1 million devices worldwide. I assume greater majority is taken up by FPGAs because Kaby-G wasn't that good nor popular.
 

jpiniero

Lifer
Oct 1, 2010
16,493
6,983
136
I wonder if the leak just doesn't show the -H models, or they are really sticking with Comet Lake for another year.

Even if you increase TDP, the -H series needs more cores than 6 and Rocketlake doesn't deliver that. Maybe that's just undecided and they'll just make the -H series by reusing the -S chips.

The rumors suggest they are just going to leave the 8 core Comet Lake-H on the market for some time. Rocket Lake might not offer much without being able to blow out the power consumption other than the updated IGP but most of those i9 laptops are aimed at high end gaming so it might not matter. Using the S line is a reasonable option if for some reason they need it.
 

inf64

Diamond Member
Mar 11, 2011
3,884
4,691
136
No they don’t. Sunny Cove is an 18% IPC increase over Skylake, which puts it about 10% above Zen 2.
That's about right. Anandtech measured around 10% difference so AMD needs 15-20% with Zen3 to get above Willow Cove which should be achievable given the latest Forrest Norrod comments and leaks.
 
Last edited:
  • Like
Reactions: spursindonesia

mikk

Diamond Member
May 15, 2012
4,291
2,381
136

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
No, Zen2 is equally to Skylake when it comes to IPC, you can see it here: https://www.overclock.net/forum/10-amd-cpus/1728758-strictly-technical-matisse-not-really.html

Maybe Zen2 has a small avantage in applications but Skylake also has a small advantage in games which overclock didn't even test.
The minute Stilt retests everything there with sunny cove included, I believe the 18% you're imagining. Till then your word on this is worth exactly as much as Intel's marketing team's, also known as the tooth fairy's word.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,865
3,729
136
No, Zen2 is equally to Skylake when it comes to IPC, you can see it here: https://www.overclock.net/forum/10-amd-cpus/1728758-strictly-technical-matisse-not-really.html

Maybe Zen2 has a small avantage in applications but Skylake also has a small advantage in games which overclock didn't even test.
That test was at a fixed, lower frequency. Andrei has gone in depth as to why testing at lower frequency artificially inflates IPC. At best, Ice Lake is 12-13 percent ahead of Zen 2 in SPEC, which is all that matters.
 
  • Like
Reactions: spursindonesia

gdansk

Diamond Member
Feb 8, 2011
4,206
7,051
136
I'm not too worried for AMD, it looks like Zen 3 may be close IPC and Intel's 10nm laptop parts, for example, are clocking lower than their 14nm parts. This is closer AMD's optimal power curve.