Discussion Intel current and future Lakes & Rapids thread

Page 723 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Saylick

Diamond Member
Sep 10, 2012
3,084
6,184
136
I think people are smarter than you think. Just because something is manufacturing ready does not mean it is in high volume manufacturing (you know like the first line states). Or you could read the last line that says "Technology readiness timing does not necessarily indicate product production timing."

Here is the actual slide if anyone wanted it:
Fair points, but odd how the slide in the article doesn't include the fine print:
1670278507549.png
 

dullard

Elite Member
May 21, 2001
24,998
3,326
126
Fair points, but odd how the slide in the article doesn't include the fine print:
It is odd that the slides are made and then later the legal people put on their legalese before the presentation? How do you do your slides, let the lawyers put on the legalese and then prepare the slides later?

Seems like you are overreaching. Please come back with substance rather than bashing messaging.
 
  • Like
Reactions: Tlh97 and ondma

dullard

Elite Member
May 21, 2001
24,998
3,326
126
Are you Intel PR now? Let me tell you something. If you have Intel stock. This is the time to sell.
Both Intel and AMD make good chips. I try to keep things as correct as possible on this forum. If anyone says anything that I see that is incorrect, whining, off subject, etc., then I like to put them in their place. This is true regardless of the company that is being discussed. After all, Intel and AMD are on the same team vs ARM in this race.

I'm fairly certain that I own both AMD and Intel stock as small fractions in my mutual funds. I do not own either one individually though. Thanks for the advice.
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
If anyone says anything that I see that is incorrect, whining, off subject, etc., then I like to put them in their place.
Are we going to pretend that Intel has been executing their product release cycle without any issues in the last 6 years?

Are you willing to bet that Meteor Lake will not be released by the end of 2023?
 
  • Like
Reactions: pcp7 and ftt

Saylick

Diamond Member
Sep 10, 2012
3,084
6,184
136
It is odd that the slides are made and then later the legal people put on their legalese before the presentation? How do you do your slides, let the lawyers put on the legalese and then prepare the slides later?

Seems like you are overreaching. Please come back with substance rather than bashing messaging.
No, you're right in that the legalese comes after the engineers put the slides together, but it doesn't make the contents of the slide any less confusing. The fact that the lawyers had to come in and add their clarifying fine print means that they thought someone might misinterpret the slides, which is a very valid concern. I mean, even Ian Cutress misinterpreted the slide on first viewing, and I know that he is a knowledgeable fellow who understands that HVM comes much later after the node is viable. It doesn't help that the slide that got distributed to the press is the version without the legalese either.

At the end of the day, I think that bashing on the messaging is valid because publicly traded companies need to be careful about what they say and how they say it. Bad messaging about their node developments, or lack thereof, is what got Intel into their predicament in the first place and they should rebuild trust with investors. Roadmaps, in the general sense, are made to convey to others what is the long-term plan, so in a way they are made to appease shareholders by telling them, "Look! We're on track for these dates!" In Intel's case, their track record as of late has been hit or miss, and they cannot afford to miss moving forward. Whether or not the slide was made to intentionally mislead, they could have done a better job at being clear.

1670280576767.png
 

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
Are we going to pretend that Intel has been executing their product release cycle without any issues in the last 6 years?

Are you willing to bet that Meteor Lake will not be released by the end of 2023?
Their client side (barring products on new nodes) have been pretty good.
Intel is able to adapt to create refreshes if they have to, in order to keep an yearly cadence on their schedule
Ex: rocket lake, which was less than impressive but was quickly followed up by the next gen product, and also raptor lake, which IS pretty impressive.
And even if they have node problems, they still seem to be able to launch low volume mobile skus as quickly as they can get the node to fab right, as seen with tiger lake. And what I'm willing to bet they will do with meteor lake.
I dislike it when people group together Intel datacenter with their client group. Given the scope of Intel's mismanagement and fab issues, what their client group has done, imo, is make the best of a bad situation and do a pretty good job. Meanwhile we have data center just shi**ng themselves every year.
 

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
No, you're right in that the legalese comes after the engineers put the slides together, but it doesn't make the contents of the slide any less confusing. The fact that the lawyers had to come in and add their clarifying fine print means that they thought someone might misinterpret the slides, which is a very valid concern. I mean, even Ian Cutress misinterpreted the slide on first viewing, and I know that he is a knowledgeable fellow who understands that HVM comes much later after the node is viable. It doesn't help that the slide that got distributed to the press is the version without the legalese either.

At the end of the day, I think that bashing on the messaging is valid because publicly traded companies need to be careful about what they say and how they say it. Bad messaging about their node developments, or lack thereof, is what got Intel into their predicament in the first place and they should rebuild trust with investors. Roadmaps, in the general sense, are made to convey to others what is the long-term plan, so in a way they are made to appease shareholders by telling them, "Look! We're on track for these dates!" In Intel's case, their track record as of late has been hit or miss, and they cannot afford to miss moving forward. Whether or not the slide was made to intentionally mislead, they could have done a better job at being clear.

View attachment 72356
It's a confusing slide for sure, but it IS heavily implied that the date for manufacturing ready isn't the date for product launch, without the text at the bottom.
I mean just look at Intel 4. It is manufacturing ready today, but we don't see Meteor Lake in HVM today do we?
That same logic should be applied for the rest of the nodes too.
It's confusing, but saying Intel is trying to mislead people with this slide is just such a bad reach.
 
  • Like
Reactions: Tlh97 and ondma

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
Fair points, but odd how the slide in the article doesn't include the fine print:
View attachment 72353
I'm really interested in that mockup for "Future Rapids" data center. Sure the Mock-up might not be fully accurate, but I still think it's really interesting how it seems like Future Rapids is using a hell of a lot more tiles than GNR, seems like they are moving to the AMD method of a bunch of small tiles rather than a couple big tiles.
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
Their client side (barring products on new nodes) have been pretty good.
Their 14nm were originally delayed(transition from 22nm) then the resorted to rehash/enhancements of that core for 5 years, They never released 10nm for the mainstream desktop... So let's not give them a free pass on Client just yet(I know about low volume Mobile 10nm SKUs)
 
  • Like
Reactions: Tlh97 and ftt

Saylick

Diamond Member
Sep 10, 2012
3,084
6,184
136
I'm really interested in that mockup for "Future Rapids" data center. Sure the Mock-up might not be fully accurate, but I still think it's really interesting how it seems like Future Rapids is using a hell of a lot more tiles than GNR, seems like they are moving to the AMD method of a bunch of small tiles rather than a couple big tiles.
I think moving to a bunch of small tiles rather than a couple of big ones is the only way forward. Cost per transistor is only going to go up, so to reduce cost for a given amount of transistors, you can 1) reduce the number of transistors needed at the cutting edge, and/or 2) reduce wasted transistors by improving yields. Chiplets accomplishes both of those goals. Of course, there is a balance point between the benefits of chiplets vs. the cost of the interconnect which is a function of many variables. Perhaps there needs to be a new "law" that says the number of chiplets in a product will double every X years or something. :p
 

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
Their 14nm were originally delayed(transition from 22nm) then the resorted to rehash/enhancements of that core for 5 years, They never released 10nm for the mainstream desktop... So let's not give them a free pass on Client just yet(I know about low volume Mobile 10nm SKUs)
Last time Intel tried to move on to a new architecture without having the new advanced node to back it up, they got rocket lake. Cores so big they had to regress in core count and IPC gains that were non-existent in applications where consumers actually cared about (cough, gaming cough). It's possible to create new architectures on the same node which have huge improvements, but they have to be planned to be built on that node in the first place.
And their client was good enough to hold the ST crown against AMD until zen 3, and they still had pretty good gen on gen improvements for being stuck on essentially the same node and arch, especially in clocks. I would call that a success.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Their client side (barring products on new nodes) have been pretty good.
Minus process related issues (which, granted, have been a huge factor), their client side have less delays than datacenter, but they've still been pretty bad. Tiger Lake should have come months earlier than it did, and more importantly, it looks like Meteor Lake has been delayed a solid year+ at this point. Irregular circumstances for MTL, perhaps, but that's no excuse.
I know about low volume Mobile 10nm SKUs
Ice Lake was one thing, but Tiger Lake probably shipped more units than AMD's entire product stack. combined.
Last time Intel tried to move on to a new architecture without having the new advanced node to back it up, they got rocket lake. Cores so big they had to regress in core count and IPC gains that were non-existent in applications where consumers actually cared about (cough, gaming cough).
RKL's big weakness was the memory subsystem, which particularly harmed gaming. The core itself seemed to be pretty much in line with where you'd expect a Sunny Cove backport to fall.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Are you willing to bet that Meteor Lake will not be released by the end of 2023?

Even if it is a paper launch or low-volume launch, Meteor Lake should appear by Q4 2023. It'll miss some key market windows like back-to-school but it'll make a showing.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
I think moving to a bunch of small tiles rather than a couple of big ones is the only way forward. Cost per transistor is only going to go up, so to reduce cost for a given amount of transistors, you can 1) reduce the number of transistors needed at the cutting edge, and/or 2) reduce wasted transistors by improving yields. Chiplets accomplishes both of those goals. Of course, there is a balance point between the benefits of chiplets vs. the cost of the interconnect which is a function of many variables. Perhaps there needs to be a new "law" that says the number of chiplets in a product will double every X years or something. :p
I don't think small chiplets for everything will become the default until hybrid bonding becomes cheap and ubiquitous, maybe another 5 years or so. The tax is just too high with any other technology. Also, cost per transistor isn't necessarily rising. Cost per area is, and for things that scale poorly (see N3 SRAM), that does mean per transistor, but logic still sees cost scaling. And of course, advanced packaging has its own monetary costs.
 
  • Like
Reactions: moinmoin

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
Minus process related issues (which, granted, have been a huge factor), their client side have less delays than datacenter, but they've still been pretty bad. Tiger Lake should have come months earlier than it did, and more importantly, it looks like Meteor Lake has been delayed a solid year+ at this point. Irregular circumstances for MTL, perhaps, but that's no excuse.

Ice Lake was one thing, but Tiger Lake probably shipped more units than AMD's entire product stack. combined.

RKL's big weakness was the memory subsystem, which particularly harmed gaming. The core itself seemed to be pretty much in line with where you'd expect a Sunny Cove backport to fall.
MTL might have been delayed a year, but RPL is already a pretty huge improvement over ADL and a worthy successor. At the very least, it's enough against Zen 4 (wins perf, loses efficiency).
TGL ramped up really well, but the first iteration did not have a high volume launch and were a little hard to find, especially when they were only launching 4-core tiger lake models.
RKL memory system was busted because (I think at least) they could not transplant Tiger Lake's memory system over to RKL. Tiger Lake had X2 the L2 and 50% more L3. And I'm pretty sure the reason they had to cut down on cache was because if they copied Tiger Lake's cache system as well, the core count would have shrunk even more than just 8.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
RPL is already a pretty huge improvement over ADL and a worthy successor. At the very least, it's enough against Zen 4 (wins perf, loses efficiency).
Raptor Lake is reasonably competitive in desktop, but Phoenix is going to slaughter it in mobile. That's where Meteor Lake's absence will be felt most keenly.
TGL ramped up really well, but the first iteration did not have a high volume launch and were a little hard to find, especially when they were only launching 4-core tiger lake models.
They launched too late in the year for high volume by the holidays. With a couple months for container ships to catch up, then things became more reasonable.
RKL memory system was busted because (I think at least) they could not transplant Tiger Lake's memory system over to RKL. Tiger Lake had X2 the L2 and 50% more L3. And I'm pretty sure the reason they had to cut down on cache was because if they copied Tiger Lake's cache system as well, the core count would have shrunk even more than just 8.
Rocket Lake wasn't just backported Sunny Cove; they backported pretty much all of Ice Lake, then tacked on Xe graphics. In retrospect, they'd probably have been better canceling it outright and using the 8c TGL die instead.
 
  • Like
Reactions: ftt

Tigerick

Senior member
Apr 1, 2022
571
446
96
It takes Intel a year to get manufacturer ready to shipping products provided that Intel don't encounter any snow stopping bugs like SPR.

So we should be expecting GNR and SF to hit OEM in H2 2024, then with qualification and testing, the servers should hit retail in 2025. Sounds about right. :p

ARL with 20A should reach OEM in H12025, we should be expecting ARL notebooks at retails before end of 2025?
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
It takes Intel a year to get manufacturer ready to shipping products provided that Intel don't encounter any snow stopping bugs like SPR.
According to what? In an ideal world where both design and process were aligned, a product would start shipping as soon as the process is ready for volume production.
 

Tigerick

Senior member
Apr 1, 2022
571
446
96
According to what? In an ideal world where both design and process were aligned, a product would start shipping as soon as the process is ready for volume production.
All new process needs to go through risk production to improve yield and identify defects. That's why TSMC normally spend a year of risk production of new process before entering HVM. So my most optimistic prediction is one year after manufactured ready of new process.

In fact, based on GNR being taped in Q2 2022, Intel should be finalized the design by Q2 2024. By then, Intel can start manufacturing GNR CPU tile, thus the timings are tight. I will not surprise that Intel push back shipping times to 2025 unless they work double times. :p
 

dullard

Elite Member
May 21, 2001
24,998
3,326
126
Are we going to pretend that Intel has been executing their product release cycle without any issues in the last 6 years?
Completely and totally irrelevant to the discussion at hand: future product release cycles. You seem to wish it was relevant, but that doesn't make it so.
Are you willing to bet that Meteor Lake will not be released by the end of 2023?
All signs point to Meteor Lake being released in 2023 at least in some form. Intel recently has been releasing some chips at the end of the year and more in January of the next year (Alder Lake and Raptor Lake). Meteor Lake has been photographed, a die shot is available, it has been powered on, they finally have EUV, Intel claims it is manufacturing ready, Intel has said it will be 2023, Intel denied the 2024 delay rumors, etc.
 
Last edited:

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
All signs point to Meteor Lake being released in 2023 at least in some form. Intel recently has been releasing some chips at the end of the year and more in January of the next year (Alder Lake and Raptor Lake). Meteor Lake has been photographed, a die shot is available, it has been powered on, they finally have EUV, Intel claims it is manufacturing ready, Intel has said it will be 2023, Intel denied the 2024 delay rumors, etc.
MTL Is in Production Ready stage.... But it's been delayed for about a year due to many issues(TSMC Tile)
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
Does anyone think the rumored Raptor Lake refresh could be an Intel 4 die shrink?
Someone in twitter(I can't remember who) posted that Intel 4 can't boost past 4 Ghz.. So I believe Intel will just use a 7+ approach(mature process) and DLVR and bring the CPU closer to their best energy efficiency curb as possible without giving too much ground to AMD.
 
  • Haha
Reactions: Tarkin77