Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.



Comparison of upcoming Intel's U-series CPU: Core Ultra 100U, Lunar Lake and Panther Lake

Model	Code-Name	Date	TDP	Node	Tiles	Main Tile	CPU	LP E-Core	LLC	GPU	Xe-cores
Core Ultra 100U	Meteor Lake	Q4 2023	15 - 57 W	Intel 4 + N5 + N6	4	tCPU	2P + 8E	2	12 MB	Intel Graphics	4
?	Lunar Lake	Q4 2024	17 - 30 W	N3B + N6	2	CPU + GPU & IMC	4P + 4E	0	12 MB	Arc	8
?	Panther Lake	Q1 2026 ?	?	Intel 18A + N3E	3	CPU + MC	4P + 8E	4	?	Arc	12

Comparison of die size of Each Tile of Meteor Lake, Arrow Lake, Lunar Lake and Panther Lake

	Meteor Lake	Arrow Lake (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	36 MB ?	12 MB	?
tCPU	66.48
tGPU	44.45
SoC	96.77
IOE	44.45
Total	252.15

Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[Exist50]

That may very well be why they appear to be going ham with GPU performance improvements going forward: they have ample capacity.

It's kind of been the opposite though. They've cut their GPU roadmap immensely. And who knows where they'll be with AI accelerators. The rest doesn't really need cutting edge lithography.

 

[SpudLobby]

So what do we know about the low power E Cores on MTL other than
- N6, separate tile from the compute tile
- Two of them
- Likely Crestmont
- Visible to the OS

Specifically:

Is this going to be A) only for idle power savings when the laptop is shut and in deep sleep - since powering down the compute tile is then possible and basic networking updates can run through modestly performing Crestmont cores on (presumably) low leakage N6 libraries?

Or B) do we have indication these can be utilized in daily, active PC use, say when browsing below some compute or interactivity threshold or reading a PDF in order to preserve power efficiency (maybe with an optional mode or autonomous, firmware use) from all the Ring/Uncore power in the compute tile - but then context switches are going to take longer and you have other processes that would have to be run.

Because being exposed in and of itself to the firmware and OS is perfectly compatible with utilizing this for A) only, but B) is where things actually get interesting and where Intel could save much more power for some simple tasks, if they have a way to pull off the migration and/or intelligent switching.

 

[eek2121]

So what do we know about the low power E Cores on MTL other than
- N6, separate tile from the compute tile
- Two of them
- Likely Crestmont
- Visible to the OS

Specifically:

Is this going to be A) only for idle power savings when the laptop is shut and in deep sleep - since powering down the compute tile is then possible and basic networking updates can run through modestly performing Crestmont cores on (presumably) low leakage N6 libraries?

Or B) do we have indication these can be utilized in daily, active PC use, say when browsing below some compute or interactivity threshold or reading a PDF in order to preserve power efficiency (maybe with an optional mode or autonomous, firmware use) from all the Ring/Uncore power in the compute tile - but then context switches are going to take longer and you have other processes that would have to be run.

Because being exposed in and of itself to the firmware and OS is perfectly compatible with utilizing this for A) only, but B) is where things actually get interesting and where Intel could save much more power for some simple tasks, if they have a way to pull off the migration and/or intelligent switching.

Honestly all we know is that the cores appear to be visible to the OS.

If Intel decides to let them be used without restriction, I can see latency being an issue that could cause regression in some workloads. However, other workloads will see a bit of a boost.

If those two cores are on the SoC or another chip that allows/encourages the CPU “chiplet” to power down, that will be pretty amazing. Everything from viewing documents to Netflix n chill will benefit. Even certain games would benefit. Shoot, I will speculate that the majority of all tasks users use their computer for could be powered by those two little cores.


Last night I watched youtube in one tab and browsed the internet in another for two hours, my machine (desktop) consistently used over 100w. A setup like that could potentially cut that down < 10W

 

[igor_kavinski]

Two cores is still too low for being usable on its own. They can only be useful when the Windows desktop is locked and there is no user activity taking place. Otherwise, the two cores will just get bogged down if all essential threads are forced to run on them.

 

[Khato]

Two cores is still too low for being usable on its own. They can only be useful when the Windows desktop is locked and there is no user activity taking place. Otherwise, the two cores will just get bogged down if all essential threads are forced to run on them.

Two cores remain perfectly adequate for the majority of non-compute usage. Effectively one core for all background tasks and the other for current user task. Outside of actual compute tasks, the only times that I recall dual core computers feeling slow compared to quad core were on startup and when silly antivirus background scans were run. Dual core will be more than adequate for the majority of web browsing and video playback (assuming GPU decode, of course.)

 

[eek2121]

Two cores remain perfectly adequate for the majority of non-compute usage. Effectively one core for all background tasks and the other for current user task. Outside of actual compute tasks, the only times that I recall dual core computers feeling slow compared to quad core were on startup and when silly antivirus background scans were run. Dual core will be more than adequate for the majority of web browsing and video playback (assuming GPU decode, of course.)

…and Microsoft defender now offloads to the GPU.

 

[nicalandia]

Is this a RaptorLake refresh? , 1P core and 4 e cores..

 

[dullard]

Is this a RaptorLake refresh? , 1P core and 4 e cores..

The U300 is Intel's cheapest bottom-of-the-barrel mobile chip launched in January 2023.

Intel® Processor U300 (8M Cache, up to 4.40 GHz) - Product Specifications | Intel

Intel® Processor U300 (8M Cache, up to 4.40 GHz) quick reference with specifications, features, and technologies.

www.intel.com

It is Raptor Lake in its worst form. Fewest cores, lowest speed, fewest GPU execution units, lowest cache, etc.

 

[poke01]

It's kind of been the opposite though. They've cut their GPU roadmap immensely. And who knows where they'll be with AI accelerators. The rest doesn't really need cutting edge lithography.

Intel first needs to get their CPUs in order. With Arrow Lake and Panther Lake the outcome will be seen. They need income rolling in and then they can start expanding again to other areas.

 

[A///]

Is this a RaptorLake refresh? , 1P core and 4 e cores..

View attachment 82438

if raptor lake refresh visitd the vet to get fixed.

 

[Exist50]

Intel first needs to get their CPUs in order. With Arrow Lake and Panther Lake the outcome will be seen. They need income rolling in and then they can start expanding again to other areas.

I don't think they have that luxury. Best case scenario, you're looking at a comeback with SRF, GNR, and PTL (ARL isn't going to be great for their financials with all the TSMC silicon). So that's 2025, maybe 2026 before anything starts showing up on the books? If they start actually investing then, then add another 3 years or so (optimistically) to get something on the market. So like 2030 before they have a serious graphics and AI offering? No, that's too late.

 

[poke01]

I don't think they have that luxury. Best case scenario, you're looking at a comeback with SRF, GNR, and PTL (ARL isn't going to be great for their financials with all the TSMC silicon). So that's 2025, maybe 2026 before anything starts showing up on the books? If they start actually investing then, then add another 3 years or so (optimistically) to get something on the market. So like 2030 before they have a serious graphics and AI offering? No, that's too late.

True, ATL won't be good financial wise for Intel but it's a needed step. it's way to ensure Intel's grasp if their own labs fail at 20A.
Not only that Intel needs to move quickly from Intel 7nm to better node.

dGPUs isn't what Intel should focus but advancing their iGPU arch from ATL and beyond. Premium Laptops will ditch Nvidia if Intel makes competent GPUs for their SoCs.Intel needs to also needs to improve their VPUs every generation as well.

The next couple of years will be interesting.

 

[Ajay]

I don't think they have that luxury. Best case scenario, you're looking at a comeback with SRF, GNR, and PTL (ARL isn't going to be great for their financials with all the TSMC silicon). So that's 2025, maybe 2026 before anything starts showing up on the books? If they start actually investing then, then add another 3 years or so (optimistically) to get something on the market. So like 2030 before they have a serious graphics and AI offering? No, that's too late.

Intel's graphics/HPC-AI roadmap seems pretty opaque to me right now.

 

[Exist50]

dGPUs isn't what Intel should focus but advancing their iGPU arch from ATL and beyond.

The two are tied at the hip. Plus there's AI, which should heavily leverage the GPU IP. At least for the server-scale stuff.

Intel's graphics/HPC-AI roadmap seems pretty opaque to me right now.

Because it's functionally non-existent beyond Battlemage. I still don't believe a word about Falcon Shores either.

 

[ashFTW]

Intel can focus on HPC/AI/ML software for the next few years. I think that’s the real differentiator with NVIDIA. Then meet it with cutting edge hardware on an advanced process down the road. If Investment options are limited, something has got to give.

 

[Geddagod]

Intel can focus on HPC/AI/ML software for the next few years. I think that’s the real differentiator with NVIDIA. Then meet it with cutting edge hardware on an advanced process down the road. If Investment options are limited, something has got to give.

I think the node thing is over stated. Intel has shown they are willing to buy TSMC wafers if their own nodes aren't good enough. I feel like the brunt of Intel's reluctance to go 'deep' into graphics is the levels of investment they have to commit to, financially and in manpower, in the design side.
If their designs were good enough, they should be perfectly willing to reserve a bunch of wafers in the future for AI GPUs, since selling those seem to be essentially just printing money.
Maybe Intel commits once again when they manage to stabilize server market share bleeding in 2024 and 2025 (hopefully).

 

[poke01]

The two are tied at the hip. Plus there's AI, which should heavily leverage the GPU IP. At least for the server-scale stuff

I would disagree. Intel does not need dGPUs, their future SoCs require massive GPU cores and if they don't have anything beyond battle mage then everything after Panter Lake will fail because Intel will be selling the whole package just like Apple is, so people expect their to improvements in every area of the SoC.

 

[Ajay]

I would disagree. Intel does not need dGPUs, their future SoCs require massive GPU cores and if they don't have anything beyond battle mage then everything after Panter Lake will fail because Intel will be selling the whole package just like Apple is, so people expect their to improvements in every area of the SoC.

If Intel doesn't have high performance AI/ML accelerators on the market going forward, they will be missing out on a huge market with great margins.

 

[poke01]

If Intel doesn't have high performance AI/ML accelerators on the market going forward, they will be missing out on a huge market with great margins.

Oh they will especially in Panther Lake. MTL will start era of AI and ML engines for Intel and Arrow Lake will massively improve on them.

 

[Ajay]

Oh they will especially in Panther Lake. MTL will start era of AI and ML engines for Intel and Arrow Lake will massively improve on them.

What? No. Big chonking thing like Nvidia's H100's and AMD's Mi300s. Not CPUs.

 

[poke01]

Intel Wants To Bring Path Tracing To Affordable GPUs & Even iGPUs With Real-Time Neural Rendering

Intel has disclosed its plans to bring Path Tracing to everyone and all kinds of GPUs with real-time neural rendering.

wccftech.com

Remember, it does not matter if Intel fails at dGPUs the future is SoCs on laptops and servers.

 

[poke01]

What? No. Big chonking thing like Nvidia's H100's and AMD's Mi300s. Not CPUs.

Isn't this a thread for consumer products? That's what ARL and MTL will be.

Intel can upscale the tech if they can get these products out as expected.

 

[A///]

atl?

atlantis lake?

 

[aigomorla]

atl?

atlantis lake?

Can someone tell me how MetorLake is MTL instead of just ML.....
MTL reminds me of Machine Translation Language.

Whose the guy that decided to put T in the middle, like how they put P in SPR for Saphire Rapids.
Can someone tell him thats not how Acronyms work.

They gonna call Raptor Lake... RPL? oh wait is it even called that?

 

[coercitiv]

Can someone tell me how MetorLake is MTL instead of just ML.....
MTL reminds me of Machine Translation Language.

And what does ML remind you of?

Here's a bunch of acronyms, notice a pattern?
SKL
KBL
CFL
CML
RKL
ADL
RPL