Discussion Intel current and future Lakes & Rapids thread

[Hitman928]

do you live in a world of numbers?

Epyc revenues grew 80mln in q3 from q2.
Intel grew DCG 1.4bln.

That is your definition of eating alive?

We've been through this before.

Intel's Q2 was unusually low in terms of both revenue and volume due to companies being concerned about the trade war and so they stockpiled parts in 4Q of 2018 which left 1H 2019 lite on sales. This stockpile cleared out by 3Q 2019, so a large increase from Q2 to Q3 was more of them returning to normal than growing. Actually, the Y/Y DCG volume shrunk by 6%.

Whereas Epyc volume grew over 50% from Q2 to Q3 and grew Y/Y as well (though no specifics were given). Yes, Intel is outselling AMD in every market (except DiY where AMD appears to be dominating), but it's obvious that AMD's share continues to grow overall and Epyc continues to increase in both revenue and volume while Intel is struggling to stop AMD's progress. Throwing their current revenue around is a straw man, no one is disputing this. It's how much AMD is growing while Intel's market share is shrinking.

If you go back and look at the timeline with Opteron, AMD is basically following the same curve, though maybe a quarter or two behind. If they continue to follow the same curve then by this time in 2021 AMD will have ~20% of the server market and then, if they don't slip up like last time, another 10% within a year.

 

[lobz]

1. The desktop market might be small but it's acutally growing.
2. > 4core chips are also heavily used in laptops (15-45W)

Even just ta 6-core Tiger-lake would address a market way larger than the 4-core ever could. It could even work for desktop gaming due-to the IPC increases (as they can already clock it to 4.4 on mobile, ~4.5 would be totally enough to compete with 9900K).

The ony reason I can explain the lack of at least a 6-core Tiger Lake on the roadmap is ... yields. Especially if you consider that they plan to make the 8-core Rocket Lake (with supposedly the same core @ 14nm). Why would they do that if they had 0 manufacturing issues?

It really is about believing more in your own sanity than in anything Intel officially states about things they know you can never check for yourself.
They are backporting an otherwise excellent new uarch to 14nm. In 2021...
This says to me that yields are not just bad, but that it's not even viable for a high volume product launch at all.
7nm to the rescue!

 

[coercitiv]

7nm to the rescue!

Actually it may be 10+++ to the rescue.

Latest arch on previous node may not be a one-time event, especially for high-performance desktop parts.

 

[lobz]

Actually it may be 10+++ to the rescue.

Latest arch on previous node may not be a one-time event, especially for high-performance desktop parts.

I remember when I saw this slide and bursted out laughing. There is only one single word I believe from the whole thing, and it's actually quite descriptive of the state intel's in for like more than a year now.
'Pathfinding'

 

[IntelUser2000]

I don't think its a functional yield issue. I believe its a parametric one.

So Tigerlake works fine in the low power envelope, but as soon as you put that in the desktops, it might have lower perf/watt than 14nm parts.

It's like how the original Ivy Bridge clocked lower than Sandy Bridge. Our resident technical expert idontcare did tests where he showed at higher frequencies Ivy Bridge used more power.

Some of us criticized their decision to sacrifice PC chips in order to boost their Atom lines(which was moot since it lost anyway). Unfortunately those decisions continued in the 14nm and 10nm processes. That's why they always talked about density.

10th gen 14nm chips is making ICL nothing more than a cool technical demo demonstrating Foveros right now. It's nice to have them, but nobody would ever really miss them.

Where are you getting that from? Icelake doesn't use Foveros. The only consumer facing product that'll use it is Lakefield.

 

[DrMrLordX]

do you live in a world of numbers?

Intel can't hide forever. But hey, feel free to ignore the benchmarks.

 

[lobz]

Where are you getting that from? Icelake doesn't use Foveros. The only consumer facing product that'll use it is Lakefield.

I totally wanted to write sunny cove, aka finally non-skylake... but I won't edit my post, so you don't have to edit yours either

 

[IntelUser2000]

The volume on Icelake isn't so low as to be a "tech demo". Intel said Icelake by end of 2019 is coming with 35 designs. By the same metric, they said Cometlake had 85.

I think Cometlake is behind that number which supports their claim of 14nm being supply constrained, but I counted 37 for Icelake, with about a dozen being super cheap devices. Some aren't even named like "HP 14". There are about 16 premium ones(such as those using LPDDR4x). Rest are DDR4 of various nature including single channel.

They sent quite a bit during the black friday season too. I speculated they were even using it to supplement lack of 14nm chips.

Whiskeylake slaps both of course. That doesn't make ICL a tech demo.

 

[krumme]

As AT testing shows even Intels own bm says 6c Cometlake is better than Icl for most task.

Intel’s Confusing Messaging: Is Comet Lake Better Than Ice Lake?

www.anandtech.com

I think Intel is lucky here in all their process mess.
What usecases and marketsegments is actually interesting for icl vs what the two lakes cant cover together?

Imo icl at that perf level is simply irrelevant. The highend business market is plenty covered by 4c wiskeylake and 6c cometlake plus gpu take highend consumer while 4c whisky can take the low end.

Intel can get 40 designwins without even having a cpu so it can never be a criteria for yield, and while they are not tech demos, in the grander sceme of Intels revenue they might as well be classified as such. Its mostly shareholder showoff and my guess is they could better skip to next gen 10nm or 7nm euv if it wasnt for the usual new model need. What is wrong with another rename? We went from 2 to 4 cores on ultrabooks after 8 years. It was a huge jump. And the most needed by far.

 

[Richie Rich]

Actually it may be 10+++ to the rescue.

Latest arch on previous node may not be a one-time event, especially for high-performance desktop parts.

Is there possibility that Intel will shift all future architectures one process node down?

• Golden Cove could arrive on 10nm in 2021 instead waiting for 7nm in 2022?
• Ocean Cove on 7nm EUV in 2022 instead of Golden?

 

[OriAr]

Is there possibility that Intel will shift all future architectures one process node down?

• Golden Cove could arrive on 10nm in 2021 instead waiting for 7nm in 2022?
• Ocean Cove on 7nm EUV in 2022 instead of Golden?

At the very least, it definitely seems like Intel is keeping its options open about doing it in the future if needed to do so, the rumor mill suggest even that Willow Cove will arrive on 14nm as Rocket Lake, so we could very well witness backporting in action already this year! I also guess as a lesson from the 10nm debacle, is that Intel now expects each ++ of a node to have better performance than the initial release of the next node, so it definitely makes that high performance parts (Such as desktop CPUs for the enthusiasts market) will be on a well refined node rather than a new one. (Which I suspect will be used for mobile at first for the energy savings).

 

[DrMrLordX]

• Golden Cove could arrive on 10nm in 2021 instead waiting for 7nm in 2022?

Most still seem to think that Alder Lake-S will be Golden Cove on 10nm.

 

[OriAr]

Most still seem to think that Alder Lake-S will be Golden Cove on 10nm.

I have a hard time imagining it being anything else, we know Willow Cove on 10nm is TGL, Ocean Cove will definitely be on 7nm, so we are pretty much left with Golden Cove

 

[Richie Rich]

Most still seem to think that Alder Lake-S will be Golden Cove on 10nm.

So Alder Lake is not Willow Cove refresh. OK, this means Golden Cove in the middle of 2021 will be serious competitor for Zen 4 in terms of IPC. If it's true Golden is much wider than ICL as suggested Keller then how much we can expect IPC increase, something like +20%?

 

[OriAr]

So Alder Lake is not Willow Cove refresh. OK, this means Golden Cove in the middle of 2021 will be serious competitor for Zen 4 in terms of IPC. If it's true Golden is much wider than ICL as suggested Keller then how much we can expect IPC increase, something like +20%?

Question is what Keller talked about when mentioning the much wider core, Golden Cove or Ocean Cove, next gen core seems to be Golden Cove but some say Jim's input will be only felt in Ocean Cove.
If it's Golden Cove then I think we are looking at a pretty big IPC increase indeed, +20% over Willow Cove doesn't sound out of place.
Intel said on their arch day a year ago that Golden Cove should have a big improvement on ST performance so we are probably looking at a big IPC increase regardless.

 

[lobz]

The volume on Icelake isn't so low as to be a "tech demo". Intel said Icelake by end of 2019 is coming with 35 designs. By the same metric, they said Cometlake had 85.

I think Cometlake is behind that number which supports their claim of 14nm being supply constrained, but I counted 37 for Icelake, with about a dozen being super cheap devices. Some aren't even named like "HP 14". There are about 16 premium ones(such as those using LPDDR4x). Rest are DDR4 of various nature including single channel.

They sent quite a bit during the black friday season too. I speculated they were even using it to supplement lack of 14nm chips.

Whiskeylake slaps both of course. That doesn't make ICL a tech demo.

I'm sorry but the number of designs will never mean to me that there is a correspondingly high volume of chips produced / shipped.

 

[OriAr]

I'm sorry but the number of designs will never mean to me that there is a correspondingly high volume of chips produced / shipped.

https://twitter.com/x/status/1208689094752120832

Intel sold $500M of reserved ICL inventory by the end of Q3, that'd represent a volume not that far from Zen 2 (That's around a million CPUs sold, maybe even more).
Considering there is roughly the number of ICL laptops Intel anticipated available on the market right now, I'd say Intel seems to ship ICL to the OEMs in time, so we are probably looking at around a million ICL units shipped to OEMs so far.
I think a million units is a bit too much to call a "Tech demo"

 

[Exist50]

Question is what Keller talked about when mentioning the much wider core, Golden Cove or Ocean Cove, next gen core seems to be Golden Cove but some say Jim's input will be only felt in Ocean Cove.
If it's Golden Cove then I think we are looking at a pretty big IPC increase indeed, +20% over Willow Cove doesn't sound out of place.
Intel said on their arch day a year ago that Golden Cove should have a big improvement on ST performance so we are probably looking at a big IPC increase regardless.

Ocean Cove is dead. Make of that what you will.

 

[OriAr]

Ocean Cove is dead. Make of that what you will.

Where did you hear that?

 

[Richie Rich]

Question is what Keller talked about when mentioning the much wider core, Golden Cove or Ocean Cove, next gen core seems to be Golden Cove but some say Jim's input will be only felt in Ocean Cove.
If it's Golden Cove then I think we are looking at a pretty big IPC increase indeed, +20% over Willow Cove doesn't sound out of place.
Intel said on their arch day a year ago that Golden Cove should have a big improvement on ST performance so we are probably looking at a big IPC increase regardless.

I agree with Jim's input timing, it takes at least 4 years to develop CPU so Ocen Cove is the first one (still limited influence as it will be just Golden Cove evolution) and most likely some post Ocean new uarch. IMHO when he arrived to Intel the Golden's Cove main uarch spec was already set and frozen. However I meant Keller could suggest Golden performace because he saw how Golden look like and data from performance simulation.

 

[lobz]

https://twitter.com/x/status/1208689094752120832

Intel sold $500M of reserved ICL inventory by the end of Q3, that'd represent a volume not that far from Zen 2 (That's around a million CPUs sold, maybe even more).
Considering there is roughly the number of ICL laptops Intel anticipated available on the market right now, I'd say Intel seems to ship ICL to the OEMs in time, so we are probably looking at around a million ICL units shipped to OEMs so far.
I think a million units is a bit too much to call a "Tech demo"

I'm sorry man, I can't really take him seriously. Please if you can find this particular piece of info somewhere else, I'm really interested and ready to seriously reconsider what I think.
At first glance, even if what he says is 100% correct, that just means intel managed to dump enough wafers to satisfy a market presence. That says still nothing about yields.
I also thought it was obvious that I didn't literally mean tech demo... sorry if I wasn't clear about that - that goes to @IntelUser2000 too.

As for ICL >= Zen2 in volume?

This would require a whole new Parody & Fantasy World subsection on AT forums.

 

[Hitman928]

https://twitter.com/x/status/1208689094752120832

Intel sold $500M of reserved ICL inventory by the end of Q3, that'd represent a volume not that far from Zen 2 (That's around a million CPUs sold, maybe even more).
Considering there is roughly the number of ICL laptops Intel anticipated available on the market right now, I'd say Intel seems to ship ICL to the OEMs in time, so we are probably looking at around a million ICL units shipped to OEMs so far.
I think a million units is a bit too much to call a "Tech demo"

Where is he getting this data?

 

[jpiniero]

You could get a million. 10k wafers at 20% yield is pretty close (~96 usable per wafer). The math does seem wrong but that's what I am getting.

 

[ondma]

https://twitter.com/x/status/1208689094752120832

Intel sold $500M of reserved ICL inventory by the end of Q3, that'd represent a volume not that far from Zen 2 (That's around a million CPUs sold, maybe even more).
Considering there is roughly the number of ICL laptops Intel anticipated available on the market right now, I'd say Intel seems to ship ICL to the OEMs in time, so we are probably looking at around a million ICL units shipped to OEMs so far.
I think a million units is a bit too much to call a "Tech demo"

Not for that poster.

 

[IntelUser2000]

I'm sorry but the number of designs will never mean to me that there is a correspondingly high volume of chips produced / shipped.

Of course not, but people buy what's available on the market. It's not available only in the Spectres and XPS of the world. Like I said you can get a DDR4 SO-DIMM single channel device that's $400 or less.

Plus the 6 core Cometlake is only available in a single SKU. Rest are equal core count comparions which puts Icelake in a favorable light.

Yes its not ideal as Icelake won't be available in a vPro version and that puts it out of business laptops, that's why you're seeing all commercial variants with Cometlake. Tigerlake won't change that either, as its also focused on client.