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

[Tigerick]

Wildcat Lake (WCL) Specs

Intel Wildcat Lake (WCL) is upcoming mobile SoC replacing Raptor Lake-U. 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 Q1 2026.

		Intel Raptor Lake U	Intel Wildcat Lake 15W?	Intel Lunar Lake		Intel Panther Lake 4+0+4
Launch Date		Q1-2024	Q2-2026	Q3-2024		Q1-2026
Model		Intel 150U	Intel Core 7	Core Ultra 7 268V		Core Ultra 7 365
Dies		2	2	2		3
Node		Intel 7 + ?	Intel 18-A + TSMC N6	TSMC N3B + N6		Intel 18-A + Intel 3 + TSMC N6
CPU		2 P-core + 8 E-cores	2 P-core + 4 LP E-cores	4 P-core + 4 LP E-cores		4 P-core + 4 LP E-cores
Threads		12	6	8		8
Max Clock		5.4 GHz	?	5 GHz		4.8 GHz
L3 Cache		12 MB		12 MB		12 MB
TDP		15 - 55 W	15 W ?	17 - 37 W		25 - 55 W
Memory		128-bit LPDDR5-5200	64-bit LPDDR5	128-bit LPDDR5x-8533		128-bit LPDDR5x-7467
Size		96 GB		32 GB		128 GB
Bandwidth				136 GB/s
GPU		Intel Graphics	Intel Graphics	Arc 140V		Intel Graphics
RT		No	No	YES		YES
EU / Xe		96 EU	2 Xe	8 Xe		4 Xe
Max Clock		1.3 GHz	?	2 GHz		2.5 GHz
NPU		GNA 3.0	18 TOPS	48 TOPS		49 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.

 

[LightningZ71]

Thank the deity of your choice that Intel canned Optane then. All those budget books with 4GB ram and a 16GB Optane being pedaled with mechanical HDDs were all horrid.

 

[adroc_thurston]

Thank the deity of your choice that Intel canned Optane then. All those budget books with 4GB ram and a 16GB Optane being pedaled with mechanical HDDs were all horrid.

3DXP was a viable memory technology for exactly 0 seconds.

 

[Magio]

32 bit windows and 4 GB Ram.

Maybe this is how we get the year of the Linux desktop, every OEM goes back to 4GB and W11 can't handle it so everyone is forced to move to Linux with a lightweight DE like XFCE.

 

[LightningZ71]

3DXP was a viable memory technology for exactly 0 seconds.

It was a solution in search of a problem. Sad part is, you can buy tiny 16GB generic 3DXP modules for real cheap, but they're almost completely useless.

 

[DavidC1]

They are there for area efficiency for MT application that can benefit from them. Intel can get you more MT compute for your dollar with the E cores, that is their reasoning.

If cost wasn't a concern they could just make a couple of huge tiles full of P cores and run themselves out of business because manufacturing cost is too high.

E cores are allowing both to happen, optimize for ultra low power or for density. Remember when they had Y class chips with 7W P core chips? It sucked. The N series using E cores outperformed them in MT, because they couldn't scale down low.

It also seems whatever optimization they made to Pantherlake reduced the gap between LPE and E cores significantly, to the point where it's mostly a toss up between the two per clock. I think Geekerwan tests showed ~5% difference? Lunarlake's LPE was substantially slower than Arrowlake's E.

It was a solution in search of a problem. Sad part is, you can buy tiny 16GB generic 3DXP modules for real cheap, but they're almost completely useless.

I still think they could have done it better. Another guy did a test of limiting Windows to 2GB RAM and had Windows on a Optane drive. He said it was usable. I know from experience it's a horrible experience on an SSD. So since it was premium either way they should have marketed as cheap RAM extension. They named it Optane Memory, but treated it as a HDD cache drive. I know partly the reason they did this because they saw only $ signs in front of their eyes charging $500 or something extra for the Optane server "Memory Drive" feature.

RAM isn't just performance, but you need enough for compatibility reasons. Then the future generations could have been literally slow RAM with the DDR versions.

 

[DavidC1]

Pantherlake:

Lunarlake:

It looks like they made noticeable performance gain, while keeping low power efficiency.

 

[adroc_thurston]

It was a solution in search of a problem. Sad part is, you can buy tiny 16GB generic 3DXP modules for real cheap, but they're almost completely useless.

Not really, no.
The problem is very real but 3DXP cost per bit can never scale low enough for it to be a viable memory.

 

[DavidC1]

You can see from the above results that E core is indeed more energy efficient below about 2.8W compared to a P core and scales down much further. It just scales lower as an LPE configuration. In Meteorlake that's true as well. Meteorlake data shows why it didn't do well as Intel hoped - there's no power advantage for LPE at all over regular E cores, I assume likely due to having to power up another tile and external interconnect. AMD results show that there's a crossover point too, but their E core is less power efficient in all other power ranges and is a pure density play.

-At 2W, energy efficiency over Raptorlake E core is 70% per watt, at 3W it's 58%, and maintains 60% advantage all the way. It only requires 2.75W for Pantherlake E to achieve same perf as Raptorlake E at 10W.
-Pantherlake's LPE is 100-120% more performance at same power. Alternatively it takes anywhere from 4x to 5x more power for Raptor E to match Pantherlake LPE. 2W PTL LPE is outperforming 9.5W Arrowlake E
-LPE is ~10-15% over E, or equal to being on a more than half a node when power normalized. Consider a full node is at 15-20%. Power normalized, it's ~50% better.

Pantherlake's LPE is more optimized for higher performance than Lunarlake. The curve is steeper so it's overall superior, but Lunarlake can reach lower absolute levels, combined with memory PHY having 40% lower power. Realistically though their engineering made a good balance. It's unlikely you'll see many cases where either platform is better in battery life. Lunarlake would be better in almost nothing scenarios, Pantherlake would be countering that under anything more than near idle.

 

[jdubs03]

https://videocardz.com/newz/samsung-galaxy-book6-panther-lake-laptops-priced-for-europe-entry-model-starts-at-e1149-and-book6-ultra-reaches-e3399

Good breakdown of what the price model looks like for Panther Lake.

 

[DavidC1]

LPE in Lunarlake is significantly slower compared to LPE in Arrowlake: https://old.chipsandcheese.com/2024/12/20/skymont-in-desktop-form-atom-unleashed/

That seems to be no longer the case in Pantherlake. C&C tested desktop Arrowlake as being 25% faster compared to laptop lunarlake. If you consider ~5% for laptop vs desktop, it means Pantherlake LPE is ~15-20% faster per clock than Lunarlake LPE. Wow, so 3.3GHz LPE in Pantherlake is outperforming 3.7GHz LPE in Lunarlake by 10%.

So here is the brief on MTL/ARL the LP-E is crestmont and are on N6 in SoC without any type of cache to help them.
In LNL/PTL they are in the compute tile on N3B/18A and has special cache Called Memory Side Cache that's for LP-E so they don't starve it's a power optimized cache just to keep power low.

The Darkmont/Skymont are exactly the same cores otherwise like physically the same I have actually matched their dishots.

This is also true for Lunarlake. They "fixed" LPE performance on Pantherlake. Perhaps the MSC is not useless anymore. I gotta say, execution-wise they delivered with Pantherlake.

 

[511]

https://videocardz.com/newz/samsung-galaxy-book6-panther-lake-laptops-priced-for-europe-entry-model-starts-at-e1149-and-book6-ultra-reaches-e3399

Good breakdown of what the price model looks like for Panther Lake.

We need more laptops tbh OEMs will charge whatever they think they can get away with a model

 

[ToTTenTranz]

The problem is very real but 3DXP cost per bit can never scale low enough for it to be a viable memory.

I would be willing to pay a premium for 3DXP in cost-per-GB, just to use as an OS drive with some additional storage for a "download scratchpad". Heck, I'm looking at buying a PCIe 3.0 PLX card with a bunch of M10 drives, just for that.

 

[LightningZ71]

I've got a box of the first gen Optane 32GB drives on my shelf. I did some testing a while back with four of them mounted in pairs on two cheap 2 x m.2 PLX cards in various configurations under windows 11. If you set them up as swap file only drives on a computer with 32GB of RAM and overload it to the point where it has to swap heavily, it's actually reasonably usable. Now, it's not threadripper fast, but, it doesn't grind to a halt like things used to do in the HDD or SATA SSD days and they don't wear out like normal SSDs can.

 

[adroc_thurston]

I would be willing to pay a premium for 3DXP in cost-per-GB

The premium was 3x the cost per bit of modern DRAM.
You sure?

 

[Khato]

The premium was 3x the cost per bit of modern DRAM.
You sure?

What? Which Optane drive was 3x the cost per bit of DRAM?

MSRP for the most expensive per bit offering I'm aware of was $1189 for the 400GB DC P5800X. That comes to $2.975/GB. Which is equivalent to $95 for a 32GB kit of DRAM.

 

[adroc_thurston]

Which Optane drive was 3x the cost per bit of DRAM?

The media was.
Optane was sold at a gigantic loss.

 

[Khato]

The media was.
Optane was sold at a gigantic loss.

As usual, please provide some manner of reliable source for such claims.

 

[511]

https://twitter.com/x/status/2024906908344201419

 

[coercitiv]

So consumer versions are highly unlikely, in the current market context they might have been a decent budget choice on DDR4 boards.

The cores themselves are technically obsolete though, NVL will be too far ahead in terms of ST perf and efficiency.