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]

207 million over the last few years. Wow.

Intel ex-CEO Gelsinger and current co-CEO slapped with lawsuit over Intel Foundry disclosures — plaintiffs demand Gelsinger surrender salary earned

LR Trust wants Pat Gelsinger and David Zinsner to return their compensation.

www.tomshardware.com

LMFAO no one would be CEO of Intel this way

I am amazed at the clownshow of the board and the shareholders they will just tank their own stock instead of actually fixing the issue clowns

 

[511]

Duplicate

 

[OneEng2]

5C will be the same as E core there will be penalty if the clock difference is 20-30%

Panter Lake is Monolithic like LNL With only GPU seperated In NVL it is similar to Clear water forest woth Hybrid Bonding

5C still has 2 virtual threads, and is identical in operation to the P cores vs ARL E Cores are drastically different than its P Cores and will perform radically different across a wide range of loads.... making scheduling very very tricky.

I agree that Panther Lake moving closer to LNL should eliminate the latency issues that are really hurting performance (IMO). I also suspect that those Lion Cove cores are more potent than they currently appear and may come alive with better IO.

 

[511]

5C still has 2 virtual threads, and is identical in operation to the P cores vs ARL E Cores are drastically different than its P Cores and will perform radically different across a wide range of loads.... making scheduling very very tricky.

You can simply spam more E cores in that area to account for the thread deficiency so 1C2T = 2C2T

 

[Markfw]

You can simply spam more E cores in that area to account for the thread deficiency so 1C2T = 2C2T

except the E-cores don't have all the same capabilities, especially in server.

 

[511]

yes no AVX-512 but it has AVX 10/256

 

[GTracing]

yes no AVX-512 but it has AVX 10/256

You got a source on that? I haven't heard anyone say that Skymont supports AVX-10. Intel's official specs for Arrow Lake don't say that AVX10 is supported.

Instruction Set Extensions: Intel® SSE4.1, Intel® SSE4.2, Intel® AVX2

Intel® Core™ Ultra 9 Processor 285K (36M Cache, up to 5.70 GHz) - Product Specifications | Intel

Intel® Core™ Ultra 9 Processor 285K (36M Cache, up to 5.70 GHz) quick reference with specifications, features, and technologies.

www.intel.com

 

[cannedlake240]

You got a source on that? I haven't heard anyone say that Skymont supports AVX-10. Intel's official specs for Arrow Lake don't say that AVX10 is supported.

Intel® Core™ Ultra 9 Processor 285K (36M Cache, up to 5.70 GHz) - Product Specifications | Intel

Intel® Core™ Ultra 9 Processor 285K (36M Cache, up to 5.70 GHz) quick reference with specifications, features, and technologies.

www.intel.com

Clearwater forest?

 

[Win2012R2]

yes no AVX-512 but it has AVX 10/256

Key question - is it double pumped like Zen 4 (so normal AVX 512 will work but at half speed) or software needs to be recompiled to specifically support 10/256? I think it's the latter, so one would be stuck in waiting for software updates.

 

[SiliconFly]

.

 

[Win2012R2]

I think it's supposed to debut with Nova Lake

Nova Lake should have final version of AVX 10.2 and APX stuff as well, before that some (server at least) CPUs should get AVX 10.1

 

[OneEng2]

You can simply spam more E cores in that area to account for the thread deficiency so 1C2T = 2C2T

Yes, but that doesn't make the scheduling any easier. It also adds more power, heat, and area to the die than does adding SMT.

Per the AVX512 discussion, I don't know that it matters much in desktop/laptop, but in server, a double pumped AVX 10.1 would still be quite inferior to a full width 512bit AVX.

I also don't think that adding a full width AVX512 path to Skymont is possible without some major design work around the entire front end and dispatch portions of the core. Once all these considerable changes were made, I think Skymont might look much more like an entirely new core than an "enhanced" Skymont. Same for adding SMT.

 

[GTracing]

Clearwater forest?

The point is that Skymont doesn't support AVX10, because if it did then Arrow lake would support it. Darkmont is rumored to be a minor update and I haven't heard anything about it supporting AVX10.

AFAIK AVX-10 and APX are rumored to come in 2026. Though I'd be happy to be proven wrong.

 

[GTracing]

I also don't think that adding a full width AVX512 path to Skymont is possible without some major design work around the entire front end and dispatch portions of the core. Once all these considerable changes were made, I think Skymont might look much more like an entirely new core than an "enhanced" Skymont. Same for adding SMT.

What is Skymont missing that would hinder it from adding AVX512?

 

[Saylick]

What is Skymont missing that would hinder it from adding AVX512?

Transistors.

 

[desrever]

LMFAO no one would be CEO of Intel this way

Nobody wants to work any more

 

[SiliconFly]

.

 

[Markfw]

They just didn't bother to add it. Although, Arctic wolf & Coyote Cove might include AVX10. Probably full 512 for server versions & 256 double pumped for clients.

This is why E-cores are useless to me. They are missing so many features and cause scheduling problems. Not worth it. And now that Intel has access to smaller nodes (theirs or TSMC) whey SHOULD be able to create powerful, efficient low power CPUs with all the features.

 

[eek2121]

Some idiot PM is in charge ofc

This is really the crux of the issue for Intel. When you are a very successful company and grow very large when you are owning the market you can afford R&D everywhere and even if only a bit of it pays off it's okay because you own the market, there is essentially no competition so you are raking in profit regardless.

But, slowly your market share erodes and instead of making the hard decisions on which projects to cut back funding, which to cancel, and which to continue. In a nutshell these are the decisions that the people at the top, specifically the CEO's over the last 15 or so years did not address correctly, at all, or quickly enough.

Now the boat has leaks everywhere. Intel needs to look at it's bread and butter and get all hands on deck for that. Everything else should be put on hold until that is rectified.

To put it in car terms let's look at BMW or Honda. They can have "pet" projects, but BMW has to get the 3 series right, Honda has to get the Accord right. They lose those and the pet projects and the panache or trickle down learning from them doesn't matter because the company is no longer relevent in the primary way it makes profit.

You have to understand that Intel is very corporate. What that means is that you will see the most absurd, nonsensical decisions come from them. Profit and Revenue drive everything, and processor performance is not a metric. Further, the company is still so used to being dominant with no competition and so large, they are STILL trying to shift course. AMD had a way easier time by virtue of dealing with this issue much earlier on. AMD ended up dropping their fabs. Intel will as well, I suspect.

Let's do a differential diagnosis on Arrow Lake. Data from TechPowerup review.
Compared to the 14900K this is where the 285K does significantly better than the 14900K.

The AI apps are over performing obviously due to the AI processor on ARL. But how about the other applications? What do each group (the ones that do better than the 14900K and the ones that do worse) have in common with one another?

AI Photo Photo Enhance +45%
AI Stable Diffusion +31%
AI NLP +18%
Blender +17%
COMSOL +16%
AV1 +15%
SHA3 +14%
NAMD +13%
V-Ray +11%
Y-Cruncher +9%

This is where it is significantly behind.
Gaming - Down about 6% on average at 1080p
Powerpoint -27%
WinRaR -24%
Premiere Pro - 18%
7 Zip decompression -18% ("only" down about 6% for compression)
Speedometer -16%
Outlook -15%
JetStream2 -14%
Avast -13%
AES -13%
MySQL -12%
Word --11%
Altium -10%
Excel -9%

Haven't checked the article, but one question I have is: are all those AI benchmarks using an NPU? NPU support requires modifications to code as well as a code recompile. New stuff won't automatically be supported.

Intel's scheduling issue could affect many of the rest, but Arrow Lake is still a turd. They could've overcome these issues.

I mentioned long ago that they should focus on e-cores and this proves it. Intel could be dropping 32 core skymont chips on an optimized process running at similar speeds with a few modifications. Single core would be equivalent, but they'd murder AMD in multicore, and realistically, that is the only area seeing decent gains these days, single cores gains are almost always small.

I actually genuinely worry about Intel. They are doing worse than most people realize. They can't layoff much more or cut much more. If things don't turn around soon, they'll be sold off either as a whole or in pieces. That hurts all of us. Shareholders must have those returns, everyone else isn't a factor.

 

[ajsdkflsdjfio]

I mentioned long ago that they should focus on e-cores and this proves it. Intel could be dropping 32 core skymont chips on an optimized process running at similar speeds with a few modifications. Single core would be equivalent, but they'd murder AMD in multicore, and realistically, that is the only area seeing decent gains these days, single cores gains are almost always small.

I agree, in a recent post I wished that Intel could release like a Skymont/Darkmont only product in the future with 16 cores with the die size of only 4-5 Lion cove cores. With a few modifications you get 90-95%(or even more) of the performance of Lion/Raptor cove with a fraction of the power and cost. Intel clearly has the potential for innovation but there is too much rot and holes to plug to focus on what they actually have.

I mean just look at the IPC here. With darkmont +5-10%, some modifications for more IPC/clock. You could be beating Zen 5 for half the die space and power. Even a year late that would be insane given you could price a 16core 9950x competitor at 200-250$ and a 32-core product equivalent to 32 zen5 cores at only 500$. I mean no wonder panther-lake has a 4-8-4 config.

Skymont in Desktop Form: Atom Unleashed

Skymont is Intel's newest E-Core architecture.

chipsandcheese.com

Also the 1P+16E gaming benchmarks in this vid make a lot more sense now given just how performant the Skymont cores can be with a minor overclock.
https://www.bilibili.com/video/BV1tW14YGEmz/

 

[eek2121]

I agree, in a recent post I wished that Intel could release like a Skymont/Darkmont only product in the future with 16 cores with the die size of only 4-5 Lion cove cores. With a few modifications you get 90-95%(or even more) of the performance of Lion/Raptor cove with a fraction of the power and cost. Intel clearly has the potential for innovation but there is too much rot and holes to plug to focus on what they actually have.
View attachment 113537View attachment 113538
I mean just look at the IPC here. With darkmont +5-10%, some modifications for more IPC/clock. You could be beating Zen 5 for half the die space and power. Even a year late that would be insane given you could price a 16core 9950x competitor at 200-250$ and a 32-core product equivalent to 32 zen5 cores at only 500$. I mean no wonder panther-lake has a 4-8-4 config.

Skymont in Desktop Form: Atom Unleashed

Skymont is Intel's newest E-Core architecture.

chipsandcheese.com

Also the 1P+16E gaming benchmarks in this vid make a lot more sense now given just how performant the Skymont cores can be with a minor overclock.
https://www.bilibili.com/video/BV1tW14YGEmz/

I was agreeing, not disagreeing. I am just saying that bizarre decisions happen for bizarre reasons. It could literally depend on overhead for each team. That is how dumb things can be. Also, it is viewed as a competition, sometimes, as you've noted, and the P-Core team has always been the all-star team since the current iteration of their architecture, so even THAT could have been a factor. Fear of loss of sales due to lack of confidence or something. They always boil it down to dollars, but the decisions never make sense. If they can improve company culture, they'll survive. Jim Keller left because he wasn't actually built for that type of mentality. I tried it once and got sucked into it once. It is a very different life that is more political than anything. One of those, I was eventually even given no job to do at all except attend meetings. Sounds like fun unless you like what you do and get bored easily. I quit later and found something more fun to do and it paid more as well. (that one job, department shifts happened, I was a new higher, got forgotten about and didn't even get sent the hardware I needed, got told to join specific meetings...got stuck in a gray area, HR probably would have figured it out eventually, but large companies move like molasses.)

 

[ajsdkflsdjfio]

I was agreeing, not disagreeing. I am just saying that bizarre decisions happen for bizarre reasons. It could literally depend on overhead for each team. That is how dumb things can be. Also, it is viewed as a competition, sometimes, as you've noted, and the P-Core team has always been the all-star team since the current iteration of their architecture, so even THAT could have been a factor. Fear of loss of sales due to lack of confidence or something. They always boil it down to dollars, but the decisions never make sense. If they can improve company culture, they'll survive. Jim Keller left because he wasn't actually built for that type of mentality. I tried it once and got sucked into it once. It is a very different life that is more political than anything. One of those, I was eventually even given no job to do at all except attend meetings. Sounds like fun unless you like what you do and get bored easily. I quit later and found something more fun to do and it paid more as well. (that one job, department shifts happened, I was a new higher, got forgotten about and didn't even get sent the hardware I needed, got told to join specific meetings...got stuck in a gray area, HR probably would have figured it out eventually, but large companies move like molasses.)

Yea I wasn't disagreeing with you either, I hadn't read the rest of your post and was just echoing your idea of Skymont-only products with more cores and/or less power/cost. Anyways yea I agree with what you are saying about company culture and the general bureaucratic bloat that comes with such large companies.

 

[DavidC1]

From C&C conclusions:

Skymont can give a good account of itself. It can’t match Zen 4 overall, but lands within striking distance more often than I’d expect from a density optimized core.

What's he talking about? Clock for clock performance is better than Zen 4 according to his SpecInt and SpecFP scores. Remember 7950X is 5.7GHz vs 4.6GHz for 285K's SKT. Yes, it's beating Zen 4 per clock in both Int and FP. 23% clock difference but 19% in Int and 18% in FP. 7950X is AMD's previous generation flagship.

285's SKT is 24% faster in Int compared to Lunarlake's SKT per clock and probably 30-40% faster in FP.

I can understand why people might be tempted to look at Skymont’s 8-wide, high clocking core, and wonder if it’s on its way to becoming a P-Core replacement. I don’t think it’s there yet.

Oh it absolutely can. It really doesn't take a lot. Darkmont adds few tiny changes and gets a significant fraction of what Lion Cove got over Golden Cove. Chester's conclusions are just weird.

In Gracemont generation, they claimed it could match Skylake in Int. In Skymont, it can slightly beat Raptorlake in FP too. In Arctic Wolf, I wouldn't be surprised if it matches Lion Cove in absolute performance, clocks included.

 

[poke01]

x86 has been "done" since the early 90's. I'll believe it when I see it. I'm not saying it won't happen. I just don't see it happening in 5 years.

Not 5 years but say in 15-20 years.

 

[ajsdkflsdjfio]

This is why E-cores are useless to me. They are missing so many features and cause scheduling problems. Not worth it. And now that Intel has access to smaller nodes (theirs or TSMC) whey SHOULD be able to create powerful, efficient low power CPUs with all the features.

You could also say by the same logic that the P-cores are also the ones causing scheduling problems. Besides, the features they miss are mostly regarding server, why would they be useless to you as a consumer? Also the e-cores are literally perfectly designed for server being extremely area efficient and low power.

I don't know enough about CPU design to analyze how hard it would be to add avx10/512 and other server oriented features, but IMO it'd be a lot easier to add features around an already excellent base design rather than try and somehow make the current bloated P-core architecture work in a power/area constrained product like those in DC products. If the point of DC CPUs is to offer 80% of the performance per core versus consumer computing while operating at less power per core and shoving a bunch of cores in one package, why not E-cores?