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.

 

[511]

Double digits could be 15% or 50%. Would you say a single digit 1t and 20% nt are disappointing?

The biggest problem is the 18% leaked gains at 250W if the gains were at 177W like previous leaked PLs it would have been fine

 

[techjunkie123]

Double digits could be 15% or 50%. Would you say a single digit 1t and 20% nt are disappointing?

I'm a little confused by why the nT gains are so low TBH. For i9 it's 8+16 vs 8+16, E cores are supposedly much improved, P cores slightly faster, and going from Intel 7 to N3B or 20 A. I guess the only impressive part would be if that +20% nT comes at -50% power compared to raptor lake?

 

[jpiniero]

I'm a little confused by why the nT gains are so low TBH.

No HT

 

[inf64]

I'm a little confused by why the nT gains are so low TBH. For i9 it's 8+16 vs 8+16, E cores are supposedly much improved, P cores slightly faster, and going from Intel 7 to N3B or 20 A. I guess the only impressive part would be if that +20% nT comes at -50% power compared to raptor lake?

8 P cores in ARL lack SMT which costs 30% of performance, while probably having a bit lower all core Turbo versus Raptor Lake P cores.

 

[techjunkie123]

No HT

8 P cores in ARL lack SMT which costs 30% of performance, while probably having a bit lower all core Turbo versus Raptor Lake P cores.

I see makes sense I guess. Then at least power draw should be much lower, aside from chiplet overhead. I guess we'll have to see when the retail units are out.

 

[511]

8 P cores in ARL lack SMT which costs 30% of performance, while probably having a bit lower all core Turbo versus Raptor Lake P cores.

It's not exactly 30% it varies by workload for some workload it's single digit

 

[SiliconFly]

-

 

[ondma]

True. It’s not exactly 30%, but upto 30% in certain workloads. In reality, it’s usually between 15% to 20%.

Yea, but we come back again to the fact that HT was supposedly removed to make ST performance better, yet the ST gains are apparently less than 10%. Gaming is the real test for me, and I fear that with the small ST gains and extra latency of chiplets, gaming may be a wash compared to RL or even a regression.

 

[techjunkie123]

Yea, but we come back again to the fact that HT was supposedly removed to make ST performance better, yet the ST gains are apparently less than 10%. Gaming is the real test for me, and I fear that with the small ST gains and extra latency of chiplets, gaming may be a wash compared to RL or even a regression.

I thought HT was removed to save area (cost)... At least on lunar lake. I guess it's cheaper/better to add skymont cores than to enable HT on the P cores.

But you would think for desktop HT would be helpful and the core design is unique anyway. So not sure why they don't include it on arrow lake.

 

[jpiniero]

I still say that HT was removed to save on validation costs.

Which as you can see... Intel is having problems with validation.

 

[ryanjagtap]

I still say that HT was removed to save on validation costs.

Which as you can see... Intel is having problems with validation.

Having two instead of three types of thread could make scheduling easier.

 

[SiliconFly]

-

 

[SiliconFly]

-

 

[mikk]

On a low power client chip ST is certainly not the most important metric. As for Lunar Lake battery life and efficiency from low to high load will be the most important. Who cares if it has 5% better or worse ST than something else.

 

[Wolverine2349]

Scheduling between P cores, E cores & LPE cores is rather difficult. Like I said before, when available, HT kicks in only when all physical cores are *already* fully saturated. HT scheduling is a lot easier compared to others.

Yes it is much easier because we have had HT and been an an SMP ecosystem with Windows NT operating systems for over 20 years now since November 2002 for HT,

E-cores and P cores heterogenous design is so new.

 

[Henry swagger]

https://twitter.com/x/status/1817762505973801383

7% behind the m3 in single.. looking good

 

[Goop_reformed]

https://twitter.com/x/status/1817762505973801383

7% behind the m3 in single.. looking good

What's the average frequency?

 

[controlflow]

https://twitter.com/x/status/1817762505973801383

7% behind the m3 in single.. looking good

Terrible LNC IPC! It needed 940783930703872 MHz to get that performance

Should be interesting to see what the actual frequencies and power looked like. I assume this part has a 30W PL2. Looks promising.
In order to be successful, LNL really needs to nail performance at the sub 15-20W range and be able to maintain actually low power while doing light to moderate activity like web surfing.

 

[511]

I wonder if we run it on Linux we can match M3 ST

 

[Henry swagger]

What's the average frequency?

4900ghz

 

[Henry swagger]

I wonder if we run it on Linux we can match M3 ST

It will i think.. snapdragon x elite performs better on linux than on Windows with geekbench 6

 

[naukkis]

Scheduling between P cores, E cores & LPE cores is rather difficult. Like I said before, when available, HT kicks in only when all physical cores are *already* fully saturated. HT scheduling is a lot easier compared to others.

It isn't. All other cores can be scheluded by their relative performance to each other - but when in Hybrid systems HT kicks in all previous scheluding need to be rechecked because HT will slow down previously fastest cores to be actually slowest ones. So when HT kicks in best performance cores will switch between non-HT reserved big cores and E-cores and that makes it pretty much impossible job to do right for best overall performance and responsivity as there will be additional overhead switching threads between best performing cores.

Scheluding hybrid cpu's where best performing cores have HT is like splitting tens on Blackjack. Most of the time it will result worse outcome so scheluder really does need to know well when splitting it's best performing cores. Best performing always working strategy is to not split them - ever. Intel was fool when they left HT on their hybrid cpu's at first place.

 

[MarkPost]

Terrible LNC IPC! It needed 940783930703872 MHz to get that performance

Should be interesting to see what the actual frequencies and power looked like. I assume this part has a 30W PL2. Looks promising.
In order to be successful, LNL really needs to nail performance at the sub 15-20W range and be able to maintain actually low power while doing light to moderate activity like web surfing.

5.0GHz average

processor_frequency": {
        "frequencies": [
            5018,
            5047,
            5071,
            5049,
            4983,
            5047,
            5009,
            4950,
            5044,
            5035,
            5068,
            5064,
            5055,
            5046,
            4906,
            4969,
            5000,
            5054,
            5054,
            4957,
            5063,
            3717,
            4639,
            4077,
            5079,
            5065,
            4964,
            5038,
            5063,
            5063,
            5069,
            5058,
            5020,
            5053,
            4961,
            4943,
            4893,
            4358,
            5000,
            5038,
            4973,
            5039,
            4610,
            4925,
            5019,
            4892,
            5052,
            5015,
            5060,
            5049

 

[Magio]

https://twitter.com/x/status/1817762505973801383

7% behind the m3 in single.. looking good

Good ST but we're still waiting for a demonstration of what the 30W SKU can truly do in MT. This is clearly a bad outlier considering we've seen the 17W PL1 parts can do 11k but I'd like to know whether LNL can do meaningfully better than that with 30W sustained.

 

[ErichZannMusician]

Good ST but we're still waiting for a demonstration of what the 30W SKU can truly do in MT. This is clearly a bad outlier considering we've seen the 17W PL1 parts can do 11k but I'd like to know whether LNL can do meaningfully better than that with 30W sustained.

The 30 W Ultra 9 288V exists solely for giving the iGPU more oomph, and will therefore be used mainly in gaming-centric designs.

Also, the Prestige 13 Meteor Lake had pretty meh cooling.

I wouldn't hold my breath for any magical MT improvements. 12-13K in GB6 MT is the upper limit of what it may achieve.

Also, there is no '30 W sustained' or, for that matter any fixed power limit in Intel Mobile CPU land unless you disable DTT using Throttlestop. So comparisons of MT score, especially on different laptop models, is meaningless.