Discussion Intel current and future Lakes & Rapids thread

[mikk]

Intel is totally screwed in the iGPU department until Meteor Lake and I bet you since that's nearly 2 years away AMD will have an answer for it too.

I don't think so. There are only 2 Rembrandt-U models and one of them has only 6CUs which is a joke. So there is only 1 model left who crushes Intel in the iGPU department.
In the overall scheme this is nothing. And by the way, Meteor Lake 2 years away, seriously? Meteor Lake is coming in H1 2023. You wil lose this bet easily.

 

[IntelUser2000]

I don't think so. There are only 2 Rembrandt-U models and one of them has only 6CUs which is a joke. So there is only 1 model left who crushes Intel in the iGPU department.
In the overall scheme this is nothing. And by the way, Meteor Lake 2 years away, seriously? Meteor Lake is coming in H1 2023. You wil lose this bet easily.

1.5 year or two doesn't make a difference because it's a long time either way. And I don't know why you have such confidence when they never released mobile early for so long. Every year it's been late fall/early winter since the Haswell generation when the releases got delayed.

Raptorlake doesn't seem to do anything on the GPU department either so it's a solid 50%+ win from AMD for 1.5 years. Tell me you really think they'll just stay with one SKU?

Even Tigerlake's iGPU win has lots of fine prints and astericks. As long as you set it at 28W, as long as you have LPDDR4X, as long as it's not artifacting due to drivers.

You are also assuming AMD will do nothing when Meteorlake comes out. Yes, Phoenix is a small iGPU gain but what about after that?

 

[Abwx]

I don't think so. There are only 2 Rembrandt-U models and one of them has only 6CUs which is a joke. So there is only 1 model left who crushes Intel in the iGPU department.

That s not set in stone, they have the option to release SKUs with 10 and 8 CUs when it suit them, according to the market trend and competitive landscape.

 

[IntelUser2000]

That s not set in stone, they have the option to release SKUs with 10 and 8 CUs when it suit them, according to the market trend and competitive landscape.

Exactly. AMD knows they have this in the bag, so they are optimizing for profits and playing Intel by using a cheap x600 SKU to go against them.

Alderlake iGPU got a small clock increase. But why not 128EUs? And then with Raptorlake an update to Xe-HPG derivative for 50% clock boost. At least that's something.

And we know from Alderlake-S that it's barely faster than Rocketlake-S iGPU. I mean 5-10%.

 

[Abwx]

Exactly. Alderlake iGPU got a small clock increase. But why not 128EUs? And then with Raptorlake an update to Xe-HPG derivative for 50% clock boost. At least that's something.

Because at some point they have to limit the silicon area, otherwise they ll end doing substancially less dies out of an already pricey waffer, and since they are going to sell mainstream GPUs that will render their dGPU offering much more appealing.

 

[IntelUser2000]

Because at some point they have to limit the silicon area, otherwise they ll end doing substancially less dies out of an already pricey waffer, and since they are going to sell mainstream GPUs that will render their dGPU offering much more appealing.

Something should be done, not zero which is this case.

AMD sells GPUs too, and they aren't holding back.

 

[Hitman928]

All I have to say at this point is that it is nice to have real competition again. It makes this hobby much more fun.

 

[mikk]

1.5 year or two doesn't make a difference because it's a long time either way. And I don't know why you have such confidence when they never released mobile early for so long. Every year it's been late fall/early winter since the Haswell generation when the releases got delayed.

I believe Raptor Lake and Meteor Lake will coexist for some time like ICL-U and CML-U/H did, once Meteor Lake is ready it will come regardless of Raptor Lake. Don't focus too much on Raptor Lake.

Meteor Lake taped in during Q2 2021. For comparison: ICL-U taped-in in Q2 2017 and finally launched in Q3 2019. When you say 2 years from now on you are basically claiming that Intels 4 node will be even more delayed than Intels broken 10nm with huge delays.

According to several sources Meteor Lake is coming in H1 2023, some say Q2 like the reddit leak and some like Raichu said it launches late Q1. Your predictions are sometimes really off. If you remember I told you desktop will be first in the ADL generation and you couldn't believe.

 

[Abwx]

Something should be done, not zero which is this case.

AMD sells GPUs too, and they aren't holding back.

This round since they had no choice after living years out of 10 CUs for Zen/Zen + reduced to 8 with Zen 2/3.

IIRC 2011 4C Llano had 400 SPs (6.25 CUs) and all BD based APUs iterations had 8 CUs.

So all these years the improvements where mainly frequency driven for AMD and GPU size + frequency for Intel but at a relatively slow pace for years untill the recent gens.

 

[IntelUser2000]

Meteor Lake taped in during Q2 2021. For comparison: ICL-U taped-in in Q2 2017 and finally launched in Q3 2019. When you say 2 years from now on you are basically claiming that Intels 4 node will be even more delayed than Intels broken 10nm with huge delays.

So according to your tape-in about Icelake, Meteorlake would be Q3 2023.

Tigerlake was supposed to be Summer 2020 too. Guess how that turned out?

Even H1 it gives AMD 1.5 years of free reign on the iGPU side. I have a bigger problem of them doing nothing than in regards to timing.

 

[Saylick]

All I have to say at this point is that it is nice to have real competition again. It makes this hobby much more fun.

True, but now the competition we face is against supply shortages...

 

[mikk]

So according to your tape-in about Icelake, Meteorlake would be Q3 2023.

You have to assume that Intel 4 will be as much delayed and slow as Intel 10nm back in 2017 til 2019, this is up to you. This just shows how hilarious your claim is saying it will come in 2 years.

Even H1 it gives AMD 1.5 years of free reign on the iGPU side. I have a bigger problem of them doing nothing than in regards to timing.

Once again, with only one model it doesn't matter really. If they would have a full lineup with 12CUs and no 6CU segmentation and no Barcelo this would have been a real issue.

 

[jpiniero]

You have to assume that Intel 4 will be as much delayed

Meteor Lake seems to be more destined to be at best a sideshow like Icelake was; and the following product is being fabbed at TSMC, sooo...

 

[jpiniero]

FWIW, here's AT's article on Intel's first 10 nm delay. Article is dated July 16, 2015. Skylake hadn't even been released yet.

AnandTech Forums: Technology, Hardware, Software, and Deals

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

www.anandtech.com

 

[mikk]

Meteor Lake seems to be more destined to be at best a sideshow like Icelake was; and the following product is being fabbed at TSMC, sooo...

In the end ICL-U had a bigger volume than AMDs entire Renoir, even if it was low volume for Intel.

 

[eek2121]

Meteor Lake should start shipping to OEMs what? H1'23? Add another quarter before designs show up, and Q3 or so is quite believable.

So according to your tape-in about Icelake, Meteorlake would be Q3 2023.

Tigerlake was supposed to be Summer 2020 too. Guess how that turned out?

Even H1 it gives AMD 1.5 years of free reign on the iGPU side. I have a bigger problem of them doing nothing than in regards to timing.

Meteor Lake launches Q2 of next year.

Tiger Lake did launch in the Summer of 2020. Specifically, it launch September 2nd, while Summer ended September 22nd.

 

[DrMrLordX]

True, but now the competition we face is against supply shortages...

Sad but true. The supply of available bleeding-edge silicon is going to be substantially lower in 2022/2023 than it was even ~6 years ago. Supplies will be tight.

Meteor Lake launches Q2 of next year.

That's a pretty big ask for Intel to get Intel 4/7nm out the door in working order just one year after finally launching products on 10ESF. After all the 14nm hiccups and the 10nm disaster, why would anyone think Intel 4 will hit the market in good order in such a short period of time?

 

[Exist50]

You have to assume that Intel 4 will be as much delayed and slow as Intel 10nm back in 2017 til 2019

And if the process is ready, but not the design? You still don't have a product.

 

[dullard]

That's a pretty big ask for Intel to get Intel 4/7nm out the door in working order just one year after finally launching products on 10ESF. After all the 14nm hiccups and the 10nm disaster, why would anyone think Intel 4 will hit the market in good order in such a short period of time?

Because Intel 4 is a much easier task than Intel 's 10 nm nodes. When Intel wanted to go to 10 nm, they overreached to an impossible density goal and at the same time (a) fired good engineers, (b) give the remaining engineers less production so fewer prototypes could be made, and (c) tried to jam as many new technologies into the new chips as could be done. This required inventing manufacturing techniques that had never been done in production quantities before such as self-aligned quadruple patterning with many different steps (any of which could destroy the chip and push yields towards zero). Finally Intel's 10 nm nodes rammed everything into one chip, so a flaw in one part might ruin the whole chip (it is all-or-nothing from a design point of view).

Intel 4 is simply adding EUV to a portion of the existing Intel 7 chip. EUV is much easier to design and optimize than pushing DUV lithography to its limits. Each manufacturing step is easier and there are fewer steps which helps increase yields. This is being done with rehired good engineers and more test runs. Finally, the chip is being split into multiple chiplets, so any flaw in one chiplet is less harmful to overall yields and not all chiplets need to be on Intel 4 (it isn't an all-or-nothing situation).

 

[eek2121]

Because Intel 4 is a much easier task than Intel 's 10 nm nodes. When Intel wanted to go to 10 nm, they overreached to an impossible density goal and at the same time (a) fired good engineers, (b) give the remaining engineers less production so fewer prototypes could be made, and (c) tried to jam as many new technologies into the new chips as could be done. This required inventing manufacturing techniques that had never been done in production quantities before such as self-aligned quadruple patterning with many different steps (any of which could destroy the chip and push yields towards zero).

Intel 4 is simply adding EUV to a portion of the existing Intel 7 chip. EUV is much easier to design and optimize than pushing DUV lithography to its limits. Each manufacturing step is easier and there are fewer steps which helps increase yields. This is being done with rehired good engineers and more test runs.

There will always be those users that are pessimistic. Every time Intel has been threatened by competition, they've stepped up their game. I don't see them slipping up with Intel 4. Intel 4 IS a shrink, however. Just not as much of one compared to 14nm -> 10nm. That being said, I don't think there will be a delay.

They already have a couple of different products in "production" on the node. They also have early prototypes of future products that were produced on the node. These products (one of which is MTL) work and are functional, so that leads me to believe they are on track. There is some speculation that MTL is supposed to be a pipe cleaner product for Intel 4. A way to ramp up production and improve yields with a shipping product. That is why I believe that MTL will ship on time. The quicker they get it out the door, the quicker they can get Intel 4 ready for future products.

 

[dullard]

There is some speculation that MTL is supposed to be a pipe cleaner product for Intel 4. A way to ramp up production and improve yields with a shipping product. That is why I believe that MTL will ship on time. The quicker they get it out the door, the quicker they can get Intel 4 ready for future products.

I too think that is a definite possibility. Meteor Lake may very well be a low-volume mobile-focused chip. Whatever Meteor Lake is, Intel is cramming products into much shorter timeframes than normal. According to Anandtech, there is only a half year between three nodes:

Intel 7	2021 H2 products
Rocket Lake	2022 H2
Intel 4	2023 H1 products
Intel 3	2023 H2 products

Intel Foundry Roadmap Update - New 18A-PT variant that enables 3D die stacking, 14A process node enablement

Smaller, faster, better.

www.anandtech.com

 

[DrMrLordX]

Because Intel 4 is a much easier task than Intel 's 10 nm nodes.

Easier than which of Intel's 10nm nodes? The original 10nm from 2017? That isn't saying much. A shrink is a shrink is a shrink, and even if Intel performs as well with Intel 4 as they did 14nm, they're still years out from high-yield products that can serve multiple market segments. A timely 2023 delivery of Meteor Lake in quantity would be Intel's best execution on a node shrink since 32nm->22nm.

Intel 4 is simply adding EUV to a portion of the existing Intel 7 chip.

Something tells me there's more to it than that.

@eek2121

Originally, Ponte Vecchio chiplets were to be the pipecleaner for Intel 4/7nm. That was cancelled when Intel moved those particular chiplets to N5. Intel 4's pipecleaner now appears to be Loihi 2.

 

[dullard]

Easier than which of Intel's 10nm nodes? The original 10nm from 2017? That isn't saying much. A shrink is a shrink is a shrink, and even if Intel performs as well with Intel 4 as they did 14nm, they're still years out from high-yield products that can serve multiple market segments. A timely 2023 delivery of Meteor Lake in quantity would be Intel's best execution on a node shrink since 32nm->22nm.

Something tells me there's more to it than that.

All 10 nm nodes were overstretches of physical limits with the resources that Intel was willing to spend.

That something is lying to you. Intel 4 isn't like the 10 nm fiasco. Intel 10 nm was a shrink plus so much more. Intel 4 is simply using EUV on the back end of line which simplifies connection designs between transistors (yes, someone has to do the work of simplifying the designs, but that is far less work than new libraries). Intel 3 is where Intel is putting new high density libraries, area reduction, etc. Intel 4 is basically nothing but using better manufacturing with equipment that is now already well established at other companies like TSMC.

 

[Hitman928]

All 10 nm nodes were overstretches of physical limits with the resources that Intel was willing to spend.

That something is lying to you. Intel 4 isn't like the 10 nm fiasco. Intel 10 nm was a shrink plus so much more. Intel 4 is simply using EUV on the back end of line which simplifies connection designs between transistors (yes, someone has to do the work of simplifying the designs, but that is far less work than new libraries). Intel 3 is where Intel is putting new high density libraries, area reduction, etc. Intel 4 is basically nothing but using better manufacturing with equipment that is now already well established at other companies like TSMC.

If Intel 4 is, as you claim, just increased EUV on the BEOL, then Intel will be further behind than I thought.

 

[eek2121]

If Intel 4 is, as you claim, just increased EUV on the BEOL, then Intel will be further behind than I thought.

It’s not. I have seen the numbers. Will have to find a link. Intel 4 is a bit larger than TSMC N3.

Still trying to find a better source for more specific numbers, but Intel 4’s original target was 251 MTr/mm2. An update in July from AnandTech shows they have backed off a bit, but it is still much more dense than TSMC N4 or Samsung: https://www.anandtech.com/show/1682...nm-3nm-20a-18a-packaging-foundry-emib-foveros

We will see where things land, of course.

EDIT: the table on that page should be pretty eye opening for those who wonder why Intel struggled with 10nm for so long.