Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

So I am leaning towards Raptor/Meteor being another Icelake like launch. This supports what Intel said way back when they said "12 month yield delay" but "6 month" product delay.

This is a difficult way of making up for the delays and catching up.

Just like 10nm they'll have another low yield process. Because of that you could say Intel 3 is a significant advance since it'll be like 10nm SF. And possibly
Arrowlake will use that. And you'd want Lunarlake to use Intel 20A.

Not sure about the rumors regarding Intel using TSMC 3nm for client chips. Because I know if a node isn't utilized for a main product it's dead. So it would indicate Intel 3 being dead or TSMC 3nm not being used for say, a compute tile. And if Intel 3 is an effectively dead node and a proper Intel 7nm part,* then you could say their whole fabrication ambitions need to be put to pasture, since there's no such thing as a big jump - just progressions.

*So if Meteor is using Intel 4 which may be akin to Icelake's low yield and low performance process, not using Intel 3, which is akin to Tigerlake's 10nm SF process makes no sense, which by then the issues will be pretty much solved.

 

[eek2121]

Below is the latest roadmap leak that I've seen (posted earlier in this thread).
View attachment 53921

A few rumors other rumors help give that image credibility, but as always rumors can be wrong and plans can change. I personally think up to Arrow Lake seems likely to be achievable with a very rapid succession of products. Past that I have doubts. And yes, this is a planned short overlap of products.

• Alder Lake. Just over 7 months after Rocket Lake. This the start of many short overlaps.
• Raptor Lake. May be late 3Q 2022 or early 4Q 2022. Maybe the September/October time frame that Intel used to do somewhat regularly.
• Meteor Lake. Intel's EUV is supposed to be up and running Q3 2022, so Meteor Lake 1H 2023 seems correct.
• Arrow Lake. Depends on TSMC's N3 capability. The last that I saw, their N3 node is supposed to be having revenue production in 1Q 2023, so Arrow Lake 2H 2023 should be doable.
• Lunar Lake. Intel claims that Intel 3 node will be used in manufactured products in 2H 2023. So, Lunar Lake in 2024 is possible
• Nova Lake. Seems the hardest to achieve. It involves multiple innovations that haven't been perfected yet. They have time on their hands, but Intel is possibly biting off more than they can chew again.
• Beyond. High-NA EUV is supposed to be ready Q3 2025, and may very well slip, so I think the chip after Nova Lake is 2026 at the earliest.

Latest roadmap has Arrow Lake launching in Q3 of 2023. Though curious, the leak I saw said Intel 4, not TSMC.

I don't see how that's possible since they've never made early mobile launches and Raptor Lake mobile seems to be a thing. We don't even have Alderlake mobile yet. If Raptor Lake mobile is going to have a short lifespan, most manufacturers are going to say no, considering we are still seeing Tigerlake based platforms being launched.

If they want overlapping launches, it would mean either Raptor or Meteor is a sidegrade in many aspects. Maybe it's possible with Meteor configurations having less cores but beefier GPU and Raptor the other way around.

The rest of Raptorlake seem "ok" but DLVR seems like a big deal for mobile.

Unless something like that happens I see Raptor mobile in Q1 2023 and Meteor in Q1 2024.

One disconnect I see is that Meteorlake doesn't sound very exciting from the rumor mill but Intel is putting it up front like there's nothing else. I have a suspicion that its a marketing tactic and Raptorlake will be a decent chip and Meteorlake, despite all it's breakthroughs are really another incremental improvement and nothing special except for maybe the iGPU.

Kinda like Icelake. Sure their first 10nm implementation but other than the GPU what was special? Extra battery life? Nope. Better performance? Not really.



Raichu has also stated there are Meteorlake variants with greater than 192EUs. That means a 320EU Arrow Lake is really not an upgrade.

Also, unless that 320EU is a Kaby-G like part I see a Arrowlake mobile on the horizon.

You are assuming we see a Raptor Lake mobile at all. Also, Meteor Lake is definitely on the table for early 2023. That hasn’t changed at all to my knowledge.

So I am leaning towards Raptor/Meteor being another Icelake like launch. This supports what Intel said way back when they said "12 month yield delay" but "6 month" product delay.

This is a difficult way of making up for the delays and catching up.

Just like 10nm they'll have another low yield process. Because of that you could say Intel 3 is a significant advance since it'll be like 10nm SF. And possibly
Arrowlake will use that. And you'd want Lunarlake to use Intel 20A.

Not sure about the rumors regarding Intel using TSMC 3nm for client chips. Because I know if a node isn't utilized for a main product it's dead. So it would indicate Intel 3 being dead or TSMC 3nm not being used for say, a compute tile. And if Intel 3 is an effectively dead node and a proper Intel 7nm part,* then you could say their whole fabrication ambitions need to be put to pasture, since there's no such thing as a big jump - just progressions.

*So if Meteor is using Intel 4 which may be akin to Icelake's low yield and low performance process, not using Intel 3, which is akin to Tigerlake's 10nm SF process makes no sense, which by then the issues will be pretty much solved.

Raptor Lake is Intel 7, unsure why you are comparing it to Ice Lake. There would be no reason to delay it. It replaces Alder Lake and uses the same node.

For Meteor Lake, Intel 4 is currently shipping production silicon, so I don’t expect delays. Meteor Lake will likely be used to help speed this process along and scale it up.

Meteor Lake features a new iteration of the mont cores.

 

[dullard]

Latest roadmap has Arrow Lake launching in Q3 of 2023. Though curious, the leak I saw said Intel 4, not TSMC.

I've seen it Intel 4 with a question mark and I've seen it as TSMC N3. At this point, I don't know which to believe.

 

[eek2121]

I've seen it Intel 4 with a question mark and I've seen it as TSMC N3. At this point, I don't know which to believe.

Could be dual sourced.

 

[repoman27]

Why is everyone here still talking like Intel is going to continue to make monolithic processors using a single process node? Raptor Lake is the last one. Meteor Lake and beyond are tile-based / Foveros and will necessarily leverage multiple manufacturing processes. Seriously, I feel like I'm taking crazy pills.

 

[repoman27]

For Meteor Lake, Intel 4 is currently shipping production silicon, so I don’t expect delays. Meteor Lake will likely be used to help speed this process along and scale it up.

Although I agree with your sentiment, according to Intel: "Loihi 2 has been fabricated with a pre-production version of the Intel 4 process." And in case you hadn't noticed, it is freaking tiny.

 

[coercitiv]

Why is everyone here still talking like Intel is going to continue to make monolithic processors using a single process node? Raptor Lake is the last one. Meteor Lake and beyond are tile-based / Foveros and will necessarily leverage multiple manufacturing processes. Seriously, I feel like I'm taking crazy pills.

Because despite it's may flaws, this place does have the great quality of challenging leaks and timelines. We may banter, we may argue, we may even end up insulting each-other as discussion heats up in a needless fashion, but one thing we don't do (anymore) is take promises for granted.

Intel moving to tiles does not end their node problems, execution does. Tiles come with their own set of problems, especially considering Intel does not want to play catch-up and is attempting to make a leap with their packaging. We've already seen some signs of what this means: Lakefield was underwhelming and expensive, Sapphire Rapids presented them with a new challenge as they wanted "monolithic" performance from tiles undergoing serious thermal stress.

Most of the folks around here will not give Intel the benefit of the doubt until they see consistent execution from them, deadline met after deadline met. Meanwhile you should do the same, even if it means changing your "medication".

 

[uzzi38]

You are assuming we see a Raptor Lake mobile at all.

It does exist, it doesn't change much is all. Meteor Lake should be more interesting.

 

[jpiniero]

Raptor Lake is Intel 7, unsure why you are comparing it to Ice Lake. There would be no reason to delay it. It replaces Alder Lake and uses the same node.

Meteor Lake would be the Icelake equivelent in that scenario.

 

[uzzi38]

Raichu has also stated there are Meteorlake variants with greater than 192EUs. That means a 320EU Arrow Lake is really not an upgrade.

Also, unless that 320EU is a Kaby-G like part I see a Arrowlake mobile on the horizon.

Doubt it'll be a KBL-G kind of part. Gen12 debuted with 1.35GHz boost clock with 96EUs, maxing out at ~15W (GPU only). With the new process node we should still see relatively sane power draw even with 320EUs and same clocks, and LPDDR5X memory support debuting with Raptor Lake should help keep it fed. No need for HBM or anything like that really.

Given they're willing to go that wide with the config, they could even push clocks if they wanted and still keep power budget with 30W, which I think is still somewhat reasonable for a maximum power draw.

 

[DrMrLordX]

I don't see how that's possible since they've never made early mobile launches and Raptor Lake mobile seems to be a thing.

Raptor Lake-P and Meteor Lake-M/P (whatever they call it) could coexist. Meteor Lake will be low volume by necessity, so it may behoove Intel to shore up their product stack with something they can manufacture in larger quantities.

 

[Exist50]

Doubt it'll be a KBL-G kind of part. Gen12 debuted with 1.35GHz boost clock with 96EUs, maxing out at ~15W (GPU only). With the new process node we should still see relatively sane power draw even with 320EUs and same clocks, and LPDDR5X memory support debuting with Raptor Lake should help keep it fed. No need for HBM or anything like that really.

Given they're willing to go that wide with the config, they could even push clocks if they wanted and still keep power budget with 30W, which I think is still somewhat reasonable for a maximum power draw.

Nah, you're not feeding 320 gen 12.7 EUs with just LPDDR5X. That's where ADM comes in. And I find the rumors that such a product is MTL... interesting.

 

[uzzi38]

Nah, you're not feeding 320 gen 12.7 EUs with just LPDDR5X. That's where ADM comes in. And I find the rumors that such a product is MTL... interesting.

Just trying to justify things here. Only thing I know of when it comes to Meteor Lake is a single number of EUs and that number is less than the 192EUs talked about, but more than current gen iGPUs. Where in the stack that lies or any other details are completely and totally unknown to me.

 

[repoman27]

Raichu has also stated there are Meteorlake variants with greater than 192EUs. That means a 320EU Arrow Lake is really not an upgrade.

What Raichu actually said was, "About iGPU, the max SKU maybe not only 192EU."

https://twitter.com/i/web/status/1461547295564992523

Raichu also has a pinned tweet regarding the use of the word "maybe":

https://twitter.com/i/web/status/1370006906660474885

Nah, you're not feeding 320 gen 12.7 EUs with just LPDDR5X. That's where ADM comes in. And I find the rumors that such a product is MTL... interesting.

I would have been right there with you until I saw the M1 Pro / M1 Max. Apple manages to feed the M1 Max using LPDDR5, granted they did go with 32 x16 channels to get there. Intel could follow suit and just add more memory channels.

 

[jpiniero]

I would have been right there with you until I saw the M1 Pro / M1 Max. Apple manages to feed the M1 Max using LPDDR5, granted they did go with 32 x16 channels to get there. Intel could follow suit and just add more memory channels.

Intel is definitely not doing that. A Kaby-G like product that includes a dGPU+GDDR6 on package would be more realistic.

 

[repoman27]

Intel is definitely not doing that. A Kaby-G like product that includes a dGPU+GDDR6 on package would be more realistic.

Intel wouldn't have to put the LPDDR on package though, they could design for memory down with four x64 devices on the logic board instead.

GDDR on the substrate is probably a nonstarter. It's not available stacked—each die is in a separate package, maximum of 2 GB / 32-bit interface per package.

edit: Also, power. For notebook platforms, the gulf between GDDR6 and LPDDR5 at equivalent bandwidths is pretty huge.

 

[LightningZ71]

With intel's focus on advanced packaging, could their packaging support a single stack of HBM2e? It offers the lowest power per Gbps throughput, and can satisfy the needs of any embedded gpu that they might want wit only one stack, keeping the entire assembly as small as possible.

 

[repoman27]

Most of the folks around here will not give Intel the benefit of the doubt until they see consistent execution from them, deadline met after deadline met. Meanwhile you should do the same, even if it means changing your "medication".

I realize I'm on the older side, but as I recall, Intel did execute like absolute clockwork for 8 years—prior to the slow, painful, 7-year trainwreck that began with 14nm. And although I may be irrationally optimistic when it comes to new technologies, it's not because I have any faith in Intel to deliver fully working products on schedule. I'm just here for the armchair engineering, not to root for any particular team.

I think we all develop our own sense of where the roadmaps are going for these companies, and then post about specific details based on that broader sense. Without sufficient context though, it's easy to totally misconstrue some of these comments. I'm always trying to look at things from a holistic standpoint, thinking about the whole product stack, release cadence, addressable market, available engineering resources, development timeline, maturity of manufacturing processes, etc. Which is why I have trouble looking at these tables of codenames assembled from multiple conflicting rumors from disparate sources and not trying to establish some sort of overall strategy.

 

[IntelUser2000]

Doubt it'll be a KBL-G kind of part. Gen12 debuted with 1.35GHz boost clock with 96EUs, maxing out at ~15W (GPU only). With the new process node we should still see relatively sane power draw even with 320EUs and same clocks, and LPDDR5X memory support debuting with Raptor Lake should help keep it fed. No need for HBM or anything like that really.

I meant Arrowlake mobile exists because of the 320EU leak. No way that is desktop.

Therefore the timeframe for some releases are suspect.

If they really want 5 new processes in 4 years, then Meteor Lake has to be early 2023 on Intel 4. Then it puts into question why Raptor Lake exists. Then Arrowlake has to be the next and using Intel 3 on some tiles because let's face it, if they don't and are using Intel 3 on an obscure low volume part, that process is effectively dead. If Intel 3 is dead forget about 20A cause it's a progression not a leap, make no mistake.

Since Intel 3 is an evolution of Intel 4, then it sounds similar to 10nm SF where it was actually a success unlike ICL 10nm. ICL 10nm = Meteor Lake Intel 4 is a possibility and maybe that's why Raptor Lake exists as an alternative and because the CPU part of Meteor is meh.

Then you have Lunar on Intel 20A and Nova on 18A.

 

[Exist50]

Since Intel 3 is an evolution of Intel 4, then it sounds similar to 10nm SF where it was actually a success unlike ICL 10nm. ICL 10nm = Meteor Lake Intel 4 is a possibility and maybe that's why Raptor Lake exists as an alternative and because the CPU part of Meteor is meh.

I'm thinking the opposite, actually. Intel 4 will get Grand RIdge, Meteor Lake, and Granite Rapids, but what's on Intel 3? Conflicting rumors about Lunar Lake, and that's it. Maybe a future server gen? I think it'll probably end up used more-so for IO dies and the like, especially as Intel's last node without the PowerVia complexity.

 

[eek2121]

I'm thinking the opposite, actually. Intel 4 will get Grand RIdge, Meteor Lake, and Granite Rapids, but what's on Intel 3? Conflicting rumors about Lunar Lake, and that's it. Maybe a future server gen? I think it'll probably end up used more-so for IO dies and the like, especially as Intel's last node without the PowerVia complexity.

I suspect Intel 4 will be a "long lived" node. Not to say that Intel will be stuck there, but innovations from all players are likely going to slow down drastically at that point. Case in point: TSMC N3.

 

[mikk]

Meteor Lake definitely seems to be another ICL-U type of release. Maybe it's only targeted for the lower power range while Raptor Lake mobile will also go into the higher power range. They won't have enough capacity for a full MTL lineup in Q2 2023 and the first 7nm/4 version might not clock as high. On the desktop side Intel goes straight to 7+/3 which should regain any clock deficits compared to 10ESF/7. And this could be very well another limited (K only) release in Q4. Of course we can't even rule out that Arrow Lake is another 7 process node launch but given that Intel claimed Intel 3 is ready to begin manufacturing products in the second half of 2023 it makes sense that Arrow will be Intel 3 based.

I meant Arrowlake mobile exists because of the 320EU leak. No way that is desktop.

The leaker said it's Arrow Lake-P so of course it's mobile.

 

[dullard]

I suspect Intel 4 will be a "long lived" node. Not to say that Intel will be stuck there, but innovations from all players are likely going to slow down drastically at that point. Case in point: TSMC N3.

Intel 3 is simply a revision of Intel 4. It is like 14+ vs 14 nm. Intel 4 is limited since Intel does not have enough EUV machines to make the whole chip with EUV. Instead, mostly the back end of line will use EUV (essentially the interconnects, wires, and vias that connect components to each other). Once Intel has more EUV machines, they will expand the use to the front end of line (transistors, gates, etc). Intel 3 will have a denser library due to this. Unless something is wrong with the library, the transition to Intel 3 is simply a matter of time to get equipment installed.

Past Intel 3 requires a lot of crossed fingers by Intel employees. A lot of new technology needs to work and work well. RibbonFET, PowerVias, etc. You could even lump in their advanced packaging plans that haven't been used much yet in actual products. Then Intel is banking on high-NA UEV which could very well get delayed. Lots of things could go wrong and Intel could get stuck at Intel 3 for a while until they are worked out.

Intel doesn't want Intel 3 to be long lived. But, Intel might be unable to move past it easily. So, I'd bet that Intel 3 is more of the long-lived node.

If they really want 5 new processes in 4 years, then Meteor Lake has to be early 2023 on Intel 4. Then it puts into question why Raptor Lake exists. Then Arrowlake has to be the next and using Intel 3 on some tiles because let's face it, if they don't and are using Intel 3 on an obscure low volume part, that process is effectively dead.

The more I think about it, the more I convince myself that Raptor Lake and Arrow Lake are the real deal and Meteor Lake is a short-lived, small volume pipe cleaner for all the various improvements.

 

[Doug S]

I suspect Intel 4 will be a "long lived" node. Not to say that Intel will be stuck there, but innovations from all players are likely going to slow down drastically at that point. Case in point: TSMC N3.

We don't yet know the reason why N3 broke TSMC's cadence. It might have been due to issues getting the process to work, or it might have been issues related to sufficient delivery of EUV scanners. After all, TSMC has been expanding N7+/N6 wafer starts which use a limited amount of EUV, and is in the process of expanding N5/N4 wafer starts which use even more EUV. N3 will reportedly use EUV for over 20 layers and they have some pretty aggressive targets even for risk production (30k wpm) let alone the first phase of mass production (105k wpm) If it is the latter ASML does look to be making progress with increasing yearly shipments so it may not be too much of a gating factor. I guess the wildcard is how heavily DRAM fabs get into EUV.

Before worrying that Intel may be "stuck" at Intel 4, how about worrying that they can actually get there in the promised timeline? They haven't exactly been good at keeping their promises for the past half dozen years, and all the while were saying they were on track and claiming just temporary delays.

 

[dullard]

Before worrying that Intel may be "stuck" at Intel 4, how about worrying that they can actually get there in the promised timeline? They haven't exactly been good at keeping their promises for the past half dozen years, and all the while were saying they were on track and claiming just temporary delays.

Intel didn't release much of anything exciting from ~2017 to early 2021. You are correct that the delays were far longer than Intel wanted or publicly stated. This was due to bad management, poor investment in necessary equipment, firing of great engineers, overconfidence, etc. That said, at what point do we say that they are back? What is the metric you are using to judge this?

Intel had 2 generation releases in the last 8 months. Raptor Lake has at least engineering samples in the wild. Meteor Lake is taped in and taped out. So, Intel has a reasonable shot of 4 CPU generations in just over 2 years. Is that enough, or does Intel need more to be back on track in your mind?