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

[Tigerick]

Wildcat Lake (WCL) Preliminary Specs

Intel Wildcat Lake (WCL) is upcoming mobile SoC replacing ADL-N. WCL consists of 2 tiles: compute tile and PCD tile. It is true single die consists of CPU, GPU and NPU that is fabbed by 18-A process. Last time I checked, PCD tile is fabbed by TSMC N6 process. They are connected through UCIe, not D2D; a first from Intel. Expecting launching in Q2/Computex 2026. In case people don't remember AlderLake-N, I have created a table below to compare the detail specs of ADL-N and WCL. Just for fun, I am throwing LNL and upcoming Mediatek D9500 SoC.

		Intel Alder Lake - N	Intel Wildcat Lake	Intel Lunar Lake		Mediatek D9500
Launch Date		Q1-2023	Q2-2026 ?	Q3-2024		Q3-2025
Model		Intel N300	?	Core Ultra 7 268V		Dimensity 9500 5G
Dies		2	2	2		1
Node		Intel 7 + ?	Intel 18-A + TSMC N6	TSMC N3B + N6		TSMC N3P
CPU		8 E-cores	2 P-core + 4 LP E-cores	4 P-core + 4 LP E-cores		C1 1+3+4
Threads		8	6	8		8
Max Clock		3.8 GHz	?	5 GHz
L3 Cache		6 MB	?	12 MB
TDP		7 W	Fanless ?	17 W		Fanless
Memory		64-bit LPDDR5-4800	64-bit LPDDR5-6800 ?	128-bit LPDDR5X-8533		64-bit LPDDR5X-10667
Size		16 GB	?	32 GB		24 GB ?
Bandwidth			~ 55 GB/s	136 GB/s		85.6 GB/s
GPU		UHD Graphics		Arc 140V		G1 Ultra
EU / Xe		32 EU	2 Xe	8 Xe		12
Max Clock		1.25 GHz		2 GHz
NPU		NA	18 TOPS	48 TOPS		100 TOPS ?

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.

 

[DokiDoki]

I wonder how what percentage of people who buy thin and light laptops get them with 64GB of RAM?

Why I won't buy an Intel Lunar Lake-powered laptop

Intel’s new mobile chips promise better performance and longer battery life, but they snub power users like me.

www.tomshardware.com

If you want a laptop with 64GB RAM, you buy a laptop with 64GB RAM.

You do not write asinine op-eds complaining about something that only supports a maximum of 32GB RAM.

 

[511]

I wonder how what percentage of people who buy thin and light laptops get them with 64GB of RAM?

Why I won't buy an Intel Lunar Lake-powered laptop

Intel’s new mobile chips promise better performance and longer battery life, but they snub power users like me.

www.tomshardware.com

It is <redacted> people buy 8GB Ram MacBook air thin and light and max memory on air is 24GB iirc LNL has 32 GB

Please refrain from using profanity in the tech section. -Moderator Shmee

 

[511]

I am sorry mod 🥲 didn't realise it

 

[Doug S]

I could count the %age on my hand. clickbait article

Its some weird sense of entitlement that people think that every product should be designed with their needs in mind. So someone announces a laptop with up to 32 GB and they whine that they want it with 64. Buy a different laptop if you want what it doesn't have! This is no different that you want a small laptop with a screen 13" or less but want a 4K display, and complain that you can't get a 4K display in that size and have to go up to nearly 16" to get that. If you have niche needs you're gonna have to face that most products won't be designed with your needs in mind.

 

[511]

32GB should be fine for the target audience also for 64GB we need a denser memory die or larger package size in LNL to accomodate it and i agree with you

 

[igor_kavinski]

Actually its ~35% on Intel. I tested it some time ago with my 13900K:

What RAM speed/timings for those results?

 

[igor_kavinski]

I could count the %age on my hand. clickbait article

Then you don't know power users. If you are buying something so expensive, better to have it maxed out rather than hitting a wall at some point in the device's life. Especially since maxing out the specs of the device will usually be a fraction of the total cost. Only the rotten fruit company forces people to sell their kidneys for more RAM.

 

[cebri1]

If you want a laptop with 64GB RAM, you buy a laptop with 64GB RAM.

You do not write asinine op-eds complaining about something that only supports a maximum of 32GB RAM.

He needs 64GB to run LLMs… at 1.4t/s

Jokes aside I think we are maybe a couple of gens away of running fairly large models on SoCs with large & cheap (&slow) memory with decent performance.

 

[poke01]

Then you don't know power users. If you are buying something so expensive, better to have it maxed out rather than hitting a wall at some point in the device's life. Especially since maxing out the specs of the device will usually be a fraction of the total cost. Only the rotten fruit company forces people to sell their kidneys for more RAM.

True but Power users don’t buy Lunar lake they buy Arrow lake H or HX with upgradeable memory. Memory can fail so it’s not good to solder it.

 

[igor_kavinski]

True but Power users don’t buy Lunar lake

They can buy 64GB Dell XPS with Snapdragon Elite X now. Lost sales for Intel.

 

[poke01]

They can buy 64GB Dell XPS with Snapdragon Elite X now. Lost sales for Intel.

touché

 

[Tup3x]

They can buy 64GB Dell XPS with Snapdragon Elite X now. Lost sales for Intel.

For power users it might be better option anyway since it has better mt performance. Assuming you don't have to worry about performance loss caused by emulation.

 

[jdubs03]

For power users it might be better option anyway since it has better mt performance. Assuming you don't have to worry about performance loss caused by emulation.

Yeah but then you have compatibility issues. Power users are probably the most likely to deal with programs that aren’t native to WoA. This is where it would be the HX 370s time to shine.

 

[SiliconFly]

.

 

[511]

From the rumours Xe3 will have 3072 bit XMX engines and 12 XMX unit with upto ~150 TOPS and the FP32 ALU count is 1536

 

[Anhiel]

Ignoring cost I don't think asking for 64GB is unreasonable.

case1 Mobile with 16GB and Windows 10: 2GB locked for W10; ~2GB for VRAM; you only have 12GB left. Upgrade to 32GB gives you 26GB of free RAM.

case2 Now with 32GB and Windows 11/12: 4GB locked for Windows(it scales with RAM amount); 16GB for Copilot; you only 12GB left not even counting VRAM allocation. In other words your 32GB upgrade gives you 0 more RAM.

And if you consider the forced move from 32b to 64b apps that will further diminish your free amount of RAM. If you want an equal upgrade gain as for case1 going the minimum is to go for 48GB.
There are tons of reason to have VM not to mention VBS and some antivirus app are running VMs. I see this as a future must have thing.

 

[SiliconFly]

.

 

[511]

IMO asking for 64GB Ram is not wrong

 

[SiliconFly]

.

 

[Anhiel]

You're wrong on so many levels. Makes me wonder...


Nope. RAM is still expensive. Otherwise, I'd prefer to have 1TB RAM.


Windows uses less than 1GB of VRAM during normal usage like browsing, office, etc. It dynamically allocates more system memory for VRAM only during GPU intensive tasks (like gaming).

You think co-pilot uses 16GB RAM? Hilarious. Thats the minimum system requirement. Not how much it uses.


Win64 requires 1 GB more RAM than Win32.

Also, 64bit apps in general do not use twice the amount of RAM compared to their 32bit counterparts. Just a tad more.


And for running VMs, we need go for platforms that are designed with such use cases in mind. Like ARL-H or AMD HX 370. Not thin and lights.


VBS is a thin virtualization layer built for kernel security, not a full blown virtual machine like you think. The difference is huge. And it does not use tons of RAM. Many Win10 machines and almost all Win11 machines use VBS by default (for many years now).

This is theoretical that's why I said ignoring cost.

Windows actually can use much less VRAM depending on the frame buffer. But most mobile have it set to 2GB. Again this is purely theory and normal cases not fringe case.

Yes, 16GB is Copilot requirement and since Windows auto locks for 2GB at this scale there's 14GB left (not counting VRAM) not much different to my generous allocation amount.

Try testing that 32b to 64b difference with something like GIMP with huge images (taking up 1GB VRAM). In my tests for 64b (v2.10) I can't get things done (as in hours) while the 32b (v2.8) It tooks only minutes. It's a significant difference eventhough RAM usage only had 1GB difference.

VBS uses 3% it's still a loss of RAM and performance.

 

[511]

This is theoretical that's why I said ignoring cost.

Windows actually can use much less VRAM depending on the frame buffer. But most mobile have it set to 2GB. Again this is purely theory and normal cases not fringe case.

i think you are wrong here cause cause AMD Reserves the Ram permanently unlike intel which only reserves like 128-256MB max everything else is shared pool which is dynamically allocated

VBS uses 3% it's still a loss of RAM and performance.

The performance loss is too much vs the ram loss in VBS sometimes we have so many amazing programs from Microsoft or our friendly OEMS and than we have our loard and savior Google chrome that hogs Ram like no other so 3% is still fine compared to them

 

[MarkPost]

What RAM speed/timings for those results?

dont remember for sure, but probably @6000 XMP timmings. Anyways, RAM speed or timmings are irrelevant in CB23

 

[Anhiel]

i think you are wrong here cause cause AMD Reserves the Ram permanently unlike intel which only reserves like 128-256MB max everything else is shared pool which is dynamically allocated


The performance loss is too much vs the ram loss in VBS sometimes we have so many amazing programs from Microsoft or our friendly OEMS and than we have our loard and savior Google chrome that hogs Ram like no other so 3% is still fine compared to them

I dunno about those differences but I do know it won't go down less than ~40 frames and up to about 256 frames; it takes whatever amount depending of your screen size.

The point of fixed allocation calculation is not about being precise but about making sure things can run under any circumstances. In other form of engineering you would call it safety margin.

Yes, considering most people use a browser all the time, then most likely a chat program like discord, all the other background apps and before you know it a lot of RAM is gone. Not even talking about apps getting less and less optimized.
Even if I just use Office suits (at least 2-3 open), likely also Adobe PDF, the amount of apps coming together is already at least 6 and that's before even starting on power using. I wouldn't say ordinary people these days don't need lots more than a decade ago.

 

[Wolverine2349]

Upon further checking my hope for more than 8 strong cores with good core to core latency is alive with Arrow Lake after all

https://images.anandtech.com/doci/21282/Intel%20Core%20Ultra%207%20155H%20Core%20to%20Core%20Latency.png

That is Core Ultra 155H which is Meteor Lake.

The Core to core latency between Cresmont cores which there are 8 of them (both in clusters of 4) have mid 30s ns consistent latency even with the cores in different 4 core cluster). So apparently at least with Meteor Lake which Arrow Lake is gonna have same latency per reports.

So with Arrow Lake, and Skymont e-cores having Raptor Cove IPC, maybe could buy the Core ultra 265K or 285K and disable Lion Cove cores and have a 12-16 Raptor Cove substitute cores with excellent and more importantly consistent core to core latency in mid 30s ms between all 12-16 Skymont cores regardless o which node they are in., And the Skymont cores run at 4.6GHz. Probably can get them overclocked to 5GHz and perfectly setup 12 P core Raptor Lake equivalent with slight lower 5GHz clock. I am excited.

 

[Wolverine2349]

Raptor Lake it seems there is a little more hit going to e-core clusters in latency where as Meteor Lake, there is none at all and in fact core to core latency even better between Crestmont cores even in different 4 core clusters *as long as on same ring as LPE cores of course are bad which desktop counterparts will not have and are excluded)

https://images.anandtech.com/doci/17601/Core%20i9-13900K%20Core%20to%20core%20latency%203.jpg

Though its a mild increase and pretty consistent even among the 4 different 4 core Gracemont clusters though not quite as good as Raptor Cove to Raptor Cove communication. And despite consistent 40s ns score to core latency GFracemont cores are Skylake IPC or weaker in certain aspects.

Consistency is key. AMD has even better core to core latency within a CCX than Intel as all numbers are in green. But once it exits a CCX/CCD, latency jumps enormously and it is bad as not consistent as a massive latency jump through IF.,

Supposedly AMD is supposed to improve it with Zen 6, but then again with Zen 5 was supposed to be much better, but cross CCD/CCX latency actually much worse than Zen 4 as with Zen 4 jumps into high 70s to 80s ns where as Zen 5 100s of ns almost like Laptop LPE e-cores of Meteor Lake which are irreverent for desktop on Intel side.

I am exited looking at the Core Ultra 155H graph and seeing consistent 30s core to core latency of the Crestmont cores even in different clusters on same ring. Figure Arrow Lake Skymont cores should be no different if not better and just as consistent. My dream chip may be there if I can disable the Lion Cove cores and use only Skymont with a mild 5GHz overclock. As long as Skymont can truly replicate Raptor Cove IPC and real world performance with HT disabled across the board all workloads with same RAM overclock and latency as well. But obviously with good stability and no degradation unlike RPL.