News Intel Confirms Meteor Lake Comes to Desktops Next Year

[usertests]

I think Meteor Lake for desktops may end up like a niche product. "Oh, you want better integrated graphics with lower power use, on a desktop? Here you go!"

If Intel launched the top mobile Meteor Lake configuration with 128 EUs and the new NPU to a desktop socket, that would be a fantastic 5700G killer.

 

[emv]

you wont be able to buy Arrow lake processors in a PC/Laptop in 2024. They will be in "risk production" but computers will not be available until Q1 2025 at earliest. That is why MTL is needed for desktop and laptop in 2024.

 

[cyrusfox]

you wont be able to buy Arrow lake processors in a PC/Laptop in 2024. They will be in "risk production" but computers will not be available until Q1 2025 at earliest. That is why MTL is needed for desktop and laptop in 2024.

1. Chip Disaggregation enables much more rapid releases, especially if tiles can be reused and only compute tile needs to be swapped out.
2. Intel is already doing public demos of Lunar lake, which is the successor of Arrow lake.

Due to both these 2 indicators, I would not be so sure about Arrow lake being absent from retail availability in2024. I would expect to be able to buy one of the the top K skews in the Oct - Dec 2024 timeframe, with the lower power variants filling out the bottom of the stack coming in early 2025, like almost all previous Intel Desktop CPU launches.

 

[bit_user]

Mainly office use but can play some less demanding games.

IIRC, the big iGPU option for Meteor Lake-P is 128 EU (up from the 96 that's held since Tiger Lake). So, that's equivalent to A380, except with slower RAM. Even that Meteor Lake with on-package LPDDR5X that they showed has only 2/3rds the bandwidth of the A380 dGPU - and it has to be shared with the CPU cores! Plus the A380 has 75 W all to itself.

So, look at what the benchmarks say about how your favorite games perform on an A380. Then, subtract probably somewhere between 15% and 40%. Even 1080p gamers might be disappointed.

Arrow Lake is rumored to have up to 384 EUs. Assuming they include a serious chunk of cache or something, that should be enough to be interesting.

 

[Order 66]

Yeah, Paul wrote up a whole series of articles on Meteor Lake covering the various aspects. I just covered the GPU part. But I know there’s a slide that talks about using an older Intel node for Foveros, then three TSMC nodes for the SOC, GPU, and IO die, and then finally using Intel 4 for the Compute die.

Intel Details Core Ultra ‘Meteor Lake’ Architecture, Launches December 14

The Intel 4 process node has Intel’s highest Day-Zero yields in a decade.

www.tomshardware.com

So the main tiles all get connected via EMIB, which appears to stack the tiles on top of a 22nm FinFET layer (now called Intel 16). It all sounds very complex and expensive as well. I can’t help but think the cost of a single Meteor Lake processor might be twice as expensive as a single RPL-R!

(ignoring cost for a second) wouldn't it make more sense to use a newer than 22nm which intel released their first CPU for in 2012? I feel like maybe they should use 14nm or 10nm.

 

[JarredWaltonGPU]

IIRC, the big iGPU option for Meteor Lake-P is 128 EU (up from the 96 that's held since Tiger Lake). So, that's equivalent to A380, except with slower RAM. Even that Meteor Lake with on-package LPDDR5X that they showed has only 2/3rds the bandwidth of the A380 dGPU - and it has to be shared with the CPU cores! Plus the A380 has 75 W all to itself.

So, look at what the benchmarks say about how your favorite games perform on an A380. Then, subtract probably somewhere between 15% and 40%. Even 1080p gamers might be disappointed.

Arrow Lake is rumored to have up to 384 EUs. Assuming they include a serious chunk of cache or something, that should be enough to be interesting.

And Intel "leaked" (via some swag at the Innovation event) that there's a second GPU with half as many Xe-cores. That will probably be the desktop chip, though Intel could always test the waters to see if there's any interest in a higher perf iGPU.

 

[bit_user]

(ignoring cost for a second) wouldn't it make more sense to use a newer than 22nm which intel released their first CPU for in 2012? I feel like maybe they should use 14nm or 10nm.

The solution should fit the problem. If the base layer is just wires, then you want to use the cheapest node that will hit the density and performance targets. Even 22 nm is probably overkill, for that.

"Intel manufactures the Foveros interposer with its low-cost and low-power-optimized 22FFL process. Intel places the four Meteor Lake tiles on top of a passive Foveros 3D interposer/base tile, and the tiles and interposer are then fused together with microbump connections that enable high-speed communications and efficient power delivery. The passive interposer doesn’t have any logic, so it doesn’t perform any type of processing work and mostly serves as a high-speed pathway for communication between tiles."
...
"Intel designed the Foveros die to be as low-cost as possible and still deliver on the company's electrical and performance goals — it's the cheapest die on the Meteor Lake package by orders of magnitude."

​

Sounds to me like using a smaller node would probably just be a waste of money.

 

[bit_user]

And Intel "leaked" (via some swag at the Innovation event) that there's a second GPU with half as many Xe-cores.

Yes, I've since heard that. Going from 32 EU -> 64 EU is still a nice bump, for desktop iGPU users (like me, in some of my home & work systems).

That will probably be the desktop chip, though Intel could always test the waters to see if there's any interest in a higher perf iGPU.

NUC-class machines will probably get the big iGPU. Thus, people who want that will still have an option.

 

[JarredWaltonGPU]

And now we have this:

Intel Clarifies Meteor Lake is Not for Desktop PCs: Not in Socketed Form

Meteor Lake is not coming to an LGA1851 socket near you.

www.tomshardware.com

"Oh, you mean those type of desktops! With sockets! Yeah, not getting Meteor Lake. Sorry about the confusion." LOL

Alternatively: If you put a laptop on a desk, that's basically a desktop, and so Meteor Lake is coming to desk tops, right?

 

[emv]

1. Chip Disaggregation enables much more rapid releases, especially if tiles can be reused and only compute tile needs to be swapped out.
2. Intel is already doing public demos of Lunar lake, which is the successor of Arrow lake.

Due to both these 2 indicators, I would not be so sure about Arrow lake being absent from retail availability in2024. I would expect to be able to buy one of the the top K skews in the Oct - Dec 2024 timeframe, with the lower power variants filling out the bottom of the stack coming in early 2025, like almost all previous Intel Desktop CPU launches.

intel did demos of meteor lake in 2022. It was in production in Dec 2022. Lets see what happens with meteor lake ramp and then it might become clear. working out the integration of the chiplets has been what caused the mtl delay. it is not easy. but lets see. If I can buy a ARL computer in 2024. then I am wrong. If Pat says "It is in the pipeline and could be released but we will launch at CES and then computers will be available "closely following CES" " then we can discuss how I predicted that over a year prior. If arrow lake ends up on Intel 3 or TSMC 3, I am sure you will agree that is a miss by intel correct?

 

[emv]

And now we have this:

Intel Clarifies Meteor Lake is Not for Desktop PCs: Not in Socketed Form

Meteor Lake is not coming to an LGA1851 socket near you.

www.tomshardware.com

"Oh, you mean those type of desktops! With sockets! Yeah, not getting Meteor Lake. Sorry about the confusion." LOL

Alternatively: If you put a laptop on a desk, that's basically a desktop, and so Meteor Lake is coming to desk tops, right?

Its not coming to desktop computers that people actually buy. It is coming to niche products that may sell 1000s of units. But we have 14th gen processors (raptor lake refresh, socket 1700) so that pretty much the same.

I actually felt bad for Intel people. Everyone in the press got so excited with the announcement that we were all wrong and we will actually have a MTL desktop. [enter short video of "NO SOUP FOR YOU" or "no whammy, no whammy, WHAMMY!"

I am SOOOO excited about Raptor lake Refresh launch

 

[bit_user]

If arrow lake ends up on Intel 3 or TSMC 3, I am sure you will agree that is a miss by intel correct?

Well, what would launch on Intel 3 wouldn't be called Arrow Lake. If that were in the cards, they'd probably already have it on their roadmap, by now.

It's pretty much only if Intel 3 is compatible with the Intel 4 design rules (which I'm really not sure about), that they could quickly respin Meteor Lake on Intel 3. Even then, something like that would take long enough that I'd expect to see it on their roadmap, by now. Backporting Arrow lake to Intel 3 would surely take a lot longer and likely push them outside of a 2024 launch window, in which case they'd probably be better off skipping it and going to Lunar Lake or whatever.

 

[bit_user]

I actually felt bad for Intel people. Everyone in the press got so excited with the announcement that we were all wrong and we will actually have a MTL desktop.

It's their own fault for not speaking carefully and clearly. Someone like an Executive VP should be no stranger to message discipline and should be especially careful to watch what she says in on-camera press interviews.

I am SOOOO excited about Raptor lake Refresh launch

Why?

 

[TerryLaze]

Why?

Because you still don't get sarcasm...

 

[bit_user]

Because you still don't get sarcasm...

Usually, people have the good sense to follow it with a /s or a 😉 emoji, because most people have learned that sarcasm doesn't work very well in plain text, on its own.

When in doubt, I'd rather err on the side that someone is being serious than to incorrectly assume they're being sarcastic. The reason being that if they say something unexpected that I dismiss as sarcastic when it wasn't, then I've missed a chance to learn something.

 

[emv]

It's their own fault for not speaking carefully and clearly. Someone like an Executive VP should be no stranger to message discipline and should be especially careful to watch what she says in on-camera press interviews.


Why?

1) 100% agree. she screwed up and wanted to give good news. live and learn (Pat has had some recent comments that will bite him when we see what is really happening)

2) I forgot the [BEGIN Sarcasm] note. Sorry

you are 100% correct all around

 

[cyrusfox]

If arrow lake ends up on Intel 3 or TSMC 3, I am sure you will agree that is a miss by intel correct?

ummm, which part of the package???
From what I have read Arrow is going to largely be the same as meteor with an updated compute core that from this slide appears to be 20Å

And if the color of the graphics tile is significant, maybe the GPU tile gets an update vs meteor.

I will only consider it a miss by Intel if Arrow lake does not come in some shape or form in 2024, most particulary in the desktop (Another cannon lake single device launch would be a clear miss). I don't care what process tech they use whether Internal or TSMC.

 

[bit_user]

ummm, which part of the package???
From what I have read Arrow is going to largely be the same as meteor with an updated compute core that from this slide appears to be 20Å

The graphics tile will be updated, right? Transitioning from TSMC N5 to N3E and I heard a rumor it's scaling up to a max of 3x as many shaders. That also suggests they'll be doing something novel with a L4 cache or memory, otherwise it seems like it would just starve.

So, taken together with the CPU die shrink to 20A, I think you really can't say it'll be that similar. Maybe the microarchitecture won't change much, but it should probably be a bigger change than Alder Lake -> Raptor Lake.

BTW, please don't write "20Å". Intel called it "20A" and it's a fake name, rather than a measure of anything real. I think you shouldn't suggest otherwise by using the proper Angstrom symbol.