Discussion Intel current and future Lakes & Rapids thread

Page 152 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,573
146
Interesting.

It also looks like Comet Lake doesn't support LPDDR4x anymore, and it only goes up to LPDDR3. As early as Comet Lake's launch, they showed support for it, but its not even in datasheets.

It does talk about it in the product brief, so maybe they just pulled it just like with the 28W version of Icelake?

Gen 12/Xe 32 EU might like the LPDDR4x as its going to be significantly faster than the UHD 620.
There's a tweakers roadmap (that leaked ages before CML-U launch) that explains the situation there

A CML-U stepping slated for early next year will brink LPDDR4X support.

It's also the same one that said 14nm+10nm GFX for RKL-U.
 
  • Like
Reactions: ikjadoon

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Their HEDT platform is even more lost. They have 10C with CML-S on LGA1200 but only 8C with RKL-S on LGA1200, the CPU is slower despite RKL-S launching 2021 while CML-S 2020? No idea what Intel is thinking.

That's not HEDT though. Cascade Lake-X is.

There's a tweakers roadmap (that leaked ages before CML-U launch) that explains the situation there

I don't believe in that roadmap as solid considering it was released early 2018.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,573
146
Their HEDT platform is even more lost. They have 10C with CML-S on LGA1200 but only 8C with RKL-S on LGA1200, the CPU is slower despite RKL-S launching 2021 while CML-S 2020? No idea what Intel is thinking.
Sharkbay explains the situation in the comments just a little.

No i9s in the mainstream RKL stack, all i9s are on a different platform.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Their HEDT platform is even more lost. They have 10C with CML-S on LGA1200 but only 8C with RKL-S on LGA1200, the CPU is slower despite RKL-S launching 2021 while CML-S 2020? No idea what Intel is thinking.

Trying to deal with the shortage?
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Who cares if its legit? Their roadmap is in flux. Like I said its released at least 2 years ago, considering the roadmap starts in early 2018.

At this point I would think the only thing you can expect is for things to get removed and not added. I wouldn't be surprised if the 10 core Comet S gets cancelled at this point due to the shortage, especially if that rumor about the 6 core die coming first is true.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Rocket Lake might be 8 cores max on desktop.

Rocket Lake on mobile looks to be a fail. Tiger Lake is going to be a far superior mobile platform, and RKL is only there to fulfill volume.

Right now Comet Lake provides better CPU performance, and in some cases the battery life is better. Icelake has a superior GPU and media, and it has better integration. So both have niches, and due to the 14nm supply issue, it seems number of laptops using them are quite balanced between the two.

Icelake-->Tigerlake: Upgrade in all fronts
Cometlake-->Rocketlake: Downgrade except for the GPU

If they go the MCM route, it'll result in more power consumption and reduce CPU performance as it doesn't really have clock headroom and the IMC is likely going to move to the GPU portion of the die. EMIB will improve things a bit, but that's all.

OEMs are probably going to ignore Rocket Lake mobile for the most part(as I expect the same will happen with 28W Tigerlake - 25W cTDPup does everything).

The possibility of using Skylake cores again recalls my back-of-mind belief that whatever they say they'll do is always true for the far future than anything coming soon. What I mean is their claim that cores are backported may only happen with some post Sunny Cove based variants, or may not happen at all.


Rocket Lake is better for desktops, and the much improved GPU will make it useful for vast majority of the market. You can use pricing and move the lineup around to compete. The problem is on the high-end, because the top core count is limited to 8.

@jpiniero If they are moving 8+ cores to HEDT it won't help them. 10 cores on HEDT will end up larger than on client because its based on their server die with extra I/O.
 
  • Like
Reactions: coercitiv

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
If they go the MCM route, it'll result in more power consumption and reduce CPU performance as it doesn't really have clock headroom and the IMC is likely going to move to the GPU portion of the die. EMIB will improve things a bit, but that's all.

It might be three chiplets actually - with an IO die which might be on 10 nm. You would have to factor in the ability to bin tighter with the smaller dies.

If they are moving 8+ cores to HEDT it won't help them. 10 cores on HEDT will end up larger than on client because its based on their server die with extra I/O.

It sounds like it would just revert back to only going up to i7 on mainstream.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It might be three chiplets actually - with an IO die which might be on 10 nm. You would have to factor in the ability to bin tighter with the smaller dies.

Yes, its like how it is today, but with extra die. Right now in notebooks you have two dies, one being the PCH, and other the CPU/GPU/IMC.

It sounds like it would just revert back to only going up to i7 on mainstream.

That's just marketing. I assume they did this in response to Ryzen. Let's say hypothetically that they didn't have the 10nm delay and were in lead. No i9 needed.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I wouldn't be surprised if the 10 core Comet S gets cancelled at this point due to the shortage, especially if that rumor about the 6 core die coming first is true.

That would be horrific. 10c Comet Lake-S gives Intel something to fight the 3700x/3800x assuming they price it correctly. Without that, they're in real trouble. Are they sacrificing that many wafers to Cooper Lake that everything else 14nm dies on the vine?

Cometlake-->Rocketlake: Downgrade except for the GPU

Not sure how Rocket is a downgrade in mobile. There weren't going to be 10c mobile parts anyway. It's more of a sidegrade. Tiger will be a nice improvement from IceLake except that it's still just 4c.

If they go the MCM route, it'll result in more power consumption and reduce CPU performance as it doesn't really have clock headroom and the IMC is likely going to move to the GPU portion of the die. EMIB will improve things a bit, but that's all.

The power savings of moving the IMC and iGPU to 10nm should offset the power usage from EMIB. Assuming they go that route.

Rocket Lake is better for desktops, and the much improved GPU will make it useful for vast majority of the market.

Considering how much fiercer is the competition on the desktop, I wouldn't be so sure about that.

@jpiniero If they are moving 8+ cores to HEDT it won't help them. 10 cores on HEDT will end up larger than on client because its based on their server die with extra I/O.

It looks like Intel HEDT is dead anyway. What are they going to sell there next? Cooper Lake? It's not like IceLake-SP will have enough dice floating around for them to sell such things to "enthusiasts". And those won't show up in any quantity until Q4 2020.
 
Last edited:

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
Rocket Lake on mobile looks to be a fail. Tiger Lake is going to be a far superior mobile platform, and RKL is only there to fulfill volume.
Rocket Lake is better for desktops, and the much improved GPU will make it useful for vast majority of the market. You can use pricing and move the lineup around to compete. The problem is on the high-end, because the top core count is limited to 8.
I think we should look at RL as a value move for both mobile and desktop, a low cost @ good performance product.

I'm not sure it will work though, Intel's product launches in the last few years have always been a year too late in terms of getting supremacy back (they were better timed for damage control though), and the trend seems to continue.

I hope TGL is far superior to ICL, at only 4 cores it better be.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I think we should look at RL as a value move for both mobile and desktop, a low cost @ good performance product.

I hope the cost part is true, but performance part is debatable. Off-die IMC using the same CPU core is going to be sidegrade at best.

I hope TGL is far superior to ICL, at only 4 cores it better be.

Early presentations about 2x performance in Sysmark 2014 in the 9W envelope suggests fanless laptops being mainstream in that generation. Right now, the gap between Y and U in single thread is 35-45%. Watch that gap close substantially. Just like Haswell made 15W mainstream.

Pure CPU performance won't move radically. But its more in line with nature of Moore's Law to create smaller, less power devices with same performance. It'll result in better laptops, which is really the goal.

The power savings of moving the IMC and iGPU to 10nm should offset the power usage from EMIB. Assuming they go that route.

That was true back in the year 2000. That's not true today. Whereas back then you could say 70% of power could be reduced due to process, nowadays it might be only 25%. Modern uarchs typically talk about "process-uarch" optimization.

That means you no longer get the benefits by a simple shrink, you need the circuit design teams to optimize for the process. You remember the Pascal presentation where it says it had circuitry optimizations to get clocks up don't you? Process is no longer the dominant force it used to be.

The reason Haswell beat Ivy Bridge to a pulp in the battery life department had little to do with process, but in implementation, architecture and optimization. The reason ARM systems have significant advantage is again more to do with the rest and not the process.

Considering how much fiercer is the competition on the desktop, I wouldn't be so sure about that.

It won't be easy but it'll be far better than on the enthusiast side. Plus it will be an advancement over their own chips while on mobile it'll be a downgrade.
 
Last edited:
  • Like
Reactions: coercitiv

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
nowadays it might be only 25%.

Wouldn't that be enough? I agree that the off-die IMC/iGPU will be a sidegrade (with the benefit of some better memory speed support, and maybe some better graphics). But I do not think the power penalty of relying on EMIB to connect the dice will be so great that a 25% reduction in power usage of the IMC and iGPU would be insufficient to offset that penalty.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
That would be horrific. 10c Comet Lake-S gives Intel something to fight the 3700x/3800x assuming they price it correctly. Without that, they're in real trouble. Are they sacrificing that many wafers to Cooper Lake that everything else 14nm dies on the vine?

Presumably they would have to release an 8C16T i7 with Coffee Lake Refresh's die if they end up not doing the 10 core Comet Lake-S.

As for Cooper Lake, the Skylake-SP XCC die is 694 mm2. Cooper would in theory be two slightly smaller dies.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
At this point I'm starting to wonder whether Intel actually made a mistake by keeping 10nm around. They can't make superior products to their old 14nm cores, and on top of that they can't make enough 14nm cores to satisfy demand. I favored Intel's decision since it helped their PR effort, but never imagined it would end up impacting 14nm capacity to such a degree.

Assuming they drop the 10-core, what exactly is Intel launching in H1 2020? (I still believe this is unlikely, but worth speculating on)
 
Last edited:

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
but never imagined it would end up impacting 14nm capacity to such a degree.

Big part is also due to AMD I guess since intel couldn't get away with yet another 4/8 $300 cpu and the mass market that was dual-core now is quad core. That is a lot more silicon they now need to ship.
 

BigDaveX

Senior member
Jun 12, 2014
440
216
116
At this point I'm starting to wonder whether Intel actually made a mistake by keeping 10nm around. They can't make superior products to their old 14nm cores, and on top of that they can't make enough 14nm cores to satisfy demand. I favored Intel's decision since it helped their PR effort, but never imagined it would end up impacting 14nm capacity to such a degree.
They'd already invested too much in their 10nm node to just drop it altogether. And at least it's now gotten to the point where it's functional enough that they can start shipping Ice Lake SKUs and start taking the heat off of their mobile 14nm chips a little.

It's kind of like how Intel's 90nm node was a complete disaster when it came to producing the high-frequency chips they wanted to ship at the time, but all they could do was persevere through it and apply the lessons they had learned to 65nm.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
It's kind of like how Intel's 90nm node was a complete disaster when it came to producing the high-frequency chips they wanted to ship at the time, but all they could do was persevere through it and apply the lessons they had learned to 65nm.

Intel 90nm was never this bad. They released likely tens of millions of CPUs based on that process. It didn't meet performance targets, but they still had good enough yields to sell a lot of Pres-hots.
 

yeshua

Member
Aug 7, 2019
166
134
86
They'd already invested too much in their 10nm node to just drop it altogether. And at least it's now gotten to the point where it's functional enough that they can start shipping Ice Lake SKUs and start taking the heat off of their mobile 14nm chips a little.

It's kind of like how Intel's 90nm node was a complete disaster when it came to producing the high-frequency chips they wanted to ship at the time, but all they could do was persevere through it and apply the lessons they had learned to 65nm.

It doesn't feel functional enough by any stretch of imagination. We are almost four months after Intel "released" Ice Lake CPUs and more than half of them, including the fastest SKU 1068G7 are still not available. In fact the only ones which are available in a noticeable volume are 1065G7 and 1035G1. There hasn't been a single ULV SKU released yet (open the link and sort by status).
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,572
136
It doesn't feel functional enough by any stretch of imagination. We are almost four months after Intel "released" Ice Lake CPUs and more than half of them, including the fastest SKU 1068G7 are still not available. In fact the only ones which are available in a noticeable volume are 1065G7 and 1035G1. There hasn't been a single ULV SKU released yet (open the link and sort by status).

Though I'm not a fan of Charlie (from semiaccurate) I think he was spot on on the supply update issue. The fact that Intel wrote the apology letter in the first place means ... if you decode it: "we don't expect to ramp up 10nm significantly enough in the coming quarters to alleviate the issue"

Especially if you put it in the context of Dell's response:
“Intel CPU shortages have worsened quarter-over-quarter the shortages are now impacting our commercial PC and premium consumer PC Q4 forecasted shipments.”

Most of the 10nm Ice-Lakes I've seen in the wild are from Dell XPS lineup. Yet they still get less and less chips every subsequent quarter. This IMO shows quite well that 10nm is not ramping up nearly as well as intel hoped a few months ago (for Q4)
 
  • Like
Reactions: coercitiv

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
They'd already invested too much in their 10nm node to just drop it altogether. And at least it's now gotten to the point where it's functional enough that they can start shipping Ice Lake SKUs and start taking the heat off of their mobile 14nm chips a little.
As someone already mentioned in the forums, this sunken cost fallacy at it's finest. (and not something to criticize Intel about, but rather something to learn from)

No matter how much they have invested, when 10nm fabs barely produce usable volumes and they still haven't been able to meet 14nm demand even after several quarters of adjustments in production, it should become clear that 10nm production not only fails to alleviate some of the pressure on 14nm, but is in fact the main cause of the shortage.

I can only imagine the pressure their 10nm teams are under.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,627
136
As someone already mentioned in the forums, this sunken cost fallacy at it's finest. (and not something to criticize Intel about, but rather something to learn from)

No matter how much they have invested, when 10nm fabs barely produce usable volumes and they still haven't been able to meet 14nm demand even after several quarters of adjustments in production, it should become clear that 10nm production not only fails to alleviate some of the pressure on 14nm, but is in fact the main cause of the shortage.

I can only imagine the pressure their 10nm teams are under.
Just 10nm? I would imagine all of Intel has high anxiety now.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
How about the 10nm wafers going to Xe? Aren't those significant enough, seeing they have to honor contract obligations and all?