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.

 

[inquiss]

It's just AVX512 with smaller opt

And no APC. The 512 bit implementation by AMD is great so maybe they don't need to go through AVX10 route. Are there literally any benefits?

 

[Joe NYC]

I wouldnt be so sure. At least I no longer think it'll be more expensive than Zen 5 / 4nm. The reason why is Huang and his damned H100/200/ Blackwell AI GPUs. He is selling an enormous quantity them at an enormous price and TSMC picked up on that a year or so ago and increased 4nm pricing. As of now, Im not too sure of any really big players besides Intel and to a lesser degree, AMD, that are buying up that 3nm production line. NV's next GPU may be on 2nm or even 18A (or both). I have come to believe that 4nm wafer price hike by TSMC is the reason that Zen 5 launched with seemingly mismatched pricing and has seen relatively muted price drops despite by most accounts, having terrible sales.

Wafer prices and quantities are negotiated in advance, and then the contract stays as signed. AMD and Intel may have come to TSMC at the same time seeking capacity. AMD found price of N3 to expensive and instead retrofitted Zen 5 to N4 (at some expense in terms of manpower and features). Intel just said fine, and is paying the higher price. Thinking that Intel is not paying substantially higher price for N3 wafers than AMD is paying for N5/N4 is just being in LaLa Land. AMD would have stayed with their original plan to use N3 for Zen 5 if that was the case.

Then comes the die size. We don't know what ARL die size is of the compute die, but given that Intel P-cores are bigger than Zen cores, the odds are that ARL P cores portion alone + its L2 and L3 are probably the same or bigger on N3 than Zen 5 CCD die (which is 70 mm2). But then, Intel also has 16 E-Cores on top of it.

Or comparing it to Metero Lake, which has roughly identical die compute die size, but has only 6P cores and 8 E Cores.

Same or larger die size of ARL compute tile is going to cost a good amount more on N3B than same or smaller die on N4P.

 

[Saylick]

I can see how this plays out already...

View attachment 108918

Ah, yes… Right on cue.

https://twitter.com/x/status/1843693688054264021

I’m really curious what he will say when he finds out the i7 and i9 K SKUs have a 250W power limit.

 

[511]

Ah, yes… Right on cue.

https://twitter.com/x/status/1843693688054264021

I’m really curious what he will say when he finds out the i7 and i9 K SKUs have a 250W power limit.

Power limit != Actual power consumption assuming PL2 is 250W it is solely dependent on PL1 for Intel's part

 

[H433x0n]

Ah, yes… Right on cue.

https://twitter.com/x/status/1843693688054264021

I’m really curious what he will say when he finds out the i7 and i9 K SKUs have a 250W power limit.

We’ve seen Lion Cove and Skymont in Lunar Lake, I get it’s not apples/apples but I doubt it magically becomes inefficient. Obviously it’s less power constrained and uncore is clocked higher but it’s still the same core and same node.

 

[gdansk]

It isn't magic, it's the inefficiency needed for the last 600 MHz.

 

[Det0x]

*double post*

 

[Det0x]

Intel Core Ultra 9 285K Breaks Past 5K Points To Become The Fastest Single-Threaded CPU At PassMark, 11% Faster Than 14900K

Intel Core Ultra 9 285K has set a record before its launch, achieving the fastest single-threaded CPU score at PassMark.

wccftech.com

Highest single core on desktop. Arrow lake looking good

Seems like this comment didnt age too well

"I used to be a adventurer like you, but then I took an Arrowlake to the knee" 🤣

 

[ondma]

So? If you want the best gaming processor on the market you still get the X3D variant. Nothing else beats the current version, anything extra is icing on the cake.

True, but the point I was making is that neither ARL or Zen5 is a significant improvement over the previous generation in gaming performance. It is still unknown if 9xxx x3D will be a significant improvement over 7800x3D in gaming.

 

[DrMrLordX]

at least there is Intel claiming 32% IPC Gracemount

Arrow Lake-N wen?

This is it for x86 i guess first Zen5% now core ultra 2% 💀

Again with the Zen5% nonsense. This is an Intel thread. You want to discuss Zen5? Go take it to one of the Zen5 threads and see how far you get with that nonsense.

This video is damm funny

This is an Intel thread . . .

 

[H433x0n]

It isn't magic, it's the inefficiency needed for the last 600 MHz.

If it’s efficient at 5.0ghz in LNL, it’s not going to be inefficient at 5.0ghz in ARL.

In an hypothetical scenario where it is equally efficient as the 9950X in the 125-175W range but has a higher PL2 are we really going to say it’s less efficient chip? Why ignore nuance and just harp on an arbitrary PL2 number?

 

[coercitiv]

It isn't magic, it's the inefficiency needed for the last 600 MHz.

There's more to it though. For example they finally have independent voltage rails for P and E cores. That does not directly translate in better efficiency, especially under load (where they could go full tilt), but in theory it can help both light threaded and multithreaded workloads.

 

[poke01]

I think we should wait for reviews. It could end up being more good than we think

 

[511]

Maybe intel fired marketing so i guess they can't even lie now in slides 🤣

 

[inf64]

I think we should wait for reviews. It could end up being more good than we think

Or as it usually goes, the official numbers are a bit inflated. So it could go the other way too

 

[Det0x]

Or as it usually goes, the official numbers are a bit inflated. So it could go the other way too

Or just look at what they compared against..

In gaming they compared vs the 9950X, not 7800X3D/7950X3D (on a pre-patched Windows 11?)


In content creation they compared vs the 7950X3d, not 9950X


In power consumption vs 14900k, not Zen4/Zen5


And that is Intels own benches with APO enabled in gaming and @ 250w powerlimit in content creation, and i would not surprise if the power consumption test was done @ "baseline profile" for Arrow Lake and "performance/extreme profile" for Raptor Lake

Ontop of 5200MT/s memoryspeed for Zen4/5, 5600MT/s for 14900k and 6400MT/s for the 285K..

I can see the writing on the wall when we get third-party, apples-to-apples benchmarks 🧐

 

[coercitiv]

I think we should wait for reviews. It could end up being more good than we think

ARL will probably have more variance than LNL in various workloads - I would expect there will be benchmarks, probably the ones less sensitive to latency, where Lion Cove will stretch it's legs. If we're "lucky" the 9% IPC figure includes gaming, this should hide productivity workloads where it might be quite a bit better. That being said, the same applies to the other team's product, so it will be interesting to watch how everything unfolds.

For now though, we still need to let the gaming steam out.

 

[MS_AT]

AVX10 is strictly worse thing than AVX-512 that Zen 5 has, it is likely inferior to Zen 4's version of AVX-512.

APX okay, that is in interesting addition, but in last 15+ years, CPU cores have developed many tricks to sidestep all the deficiencies that APX seeks to correct, so there won't be some large benefit from it, IMHO.

Maybe to expand a bit. AVX10 will be an improvement for Intel from the feature point of view (new instructions, masking), but the consumer version (AVX10/256) will not be an improvement from throughput point of view. Compared to Zen4 double pumped units, it will also not help with code footprint, but the new features will definitely help.

For the APX, I would love for it to be already a thing as I am running out of GPRs to hold addresses for SIMD So having 32 of them would require less gymnastics.

But for both, APX and AVX10 the real problem is that you need code compiled for them. It's the same story as with Zen5, it has the best known AVX512 implementation, but in reality there is little software making use of it.

 

[Kocicak]

I wonder if Arrow lake performs like this because it should be this way, or if it is somehow broken but it would require such a long time or expense to fix it, that Intel simply gave up and released it as is.

I wonder if the problem is not related to the disaggregation of the CPU. There are too many pieces in the puzzle and each of them needs to be 100% in order. This must be a nightmare to make.

 

[KompuKare]

Or just look at what they compared against..

In gaming they compared vs the 9950X, not 7800X3D/7950X3D
View attachment 108979

In content creation they compared vs the 7950X3d, not 9950X
View attachment 108980

In power consumption vs 14900k, not Zen4/Zen5
View attachment 108978

And that is Intels own benches with APO enabled in gaming and @ 250w powerlimit in content creation, and i would not surprise if the power consumption test was done @ "baseline profile" for Arrow Lake and "performance/extreme profile" for Raptor Lake

Ontop of 5200MT/s memoryspeed for Zen4/5, 5600MT/s for 14900k and 6400MT/s for the 285K..

I can see the writing on the wall when we get third-party, apples-to-apples benchmarks 🧐
View attachment 108981

Good points!

Internally marketing should have the full set of results from their technical marketing labs therefore whatever they actually choose to show tends to be whatever makes them look best. Sure that is marketing's job but reading between the lines is very important.

Sometime next month after all the reviews, should we create a
"2024 CPUs, their hype-train, and subsequent derailment"
in a vendor neutral thread with ARL, Zen5, SnapDragon X, etc.!?

 

[511]

MLID says LNL is not good that dude is something else 🤣

 

[Joe NYC]

Or just look at what they compared against..

In gaming they compared vs the 9950X, not 7800X3D/7950X3D (on a pre-patched Windows 11?)
View attachment 108979

In content creation they compared vs the 7950X3d, not 9950X
View attachment 108980

In power consumption vs 14900k, not Zen4/Zen5
View attachment 108978

And that is Intels own benches with APO enabled in gaming and @ 250w powerlimit in content creation, and i would not surprise if the power consumption test was done @ "baseline profile" for Arrow Lake and "performance/extreme profile" for Raptor Lake

Ontop of 5200MT/s memoryspeed for Zen4/5, 5600MT/s for 14900k and 6400MT/s for the 285K..

I can see the writing on the wall when we get third-party, apples-to-apples benchmarks 🧐
View attachment 108981

Yes, the slides don't compare against the best CPU from the competition for the category of applications, but they pick the CPU that is not the best fit.

Also, someone mentioned Intel used DDR5-6400 memory for AMD. I seriously doubt that Intel is using it in 1:1 clock ratio, that is ideal for AMD processors. AMD recommends DDR5-6000 and there are tons out there that are cheap and supporting good memory timings.

BTW, this guy on youtube goes over the benches:

 

[511]

Yes, the slides don't compare against the best CPU from the competition for the category of applications, but they pick the CPU that is not the best fit.

Also, someone mentioned Intel used DDR5-6400 memory for AMD. I seriously doubt that Intel is using it in 1:1 clock ratio, that is ideal for AMD processors. AMD recommends DDR5-6000 and there are tons out there that are cheap and supporting good memory timings.

BTW, this guy on youtube goes over the benches:

Are all slides leaked though that is my point we know how is it with the leaks we will know in 30 hours

 

[vanplayer]

BTW, this guy on youtube goes over the benches:

I love this high rated comment under this video:

Arrow-in-the-knee Lake

 

[tsamolotoff]

This person seems happy

Another schizo from WCCF like the previously mentioned guy with latina nickname (Jubileu?), Indian guy and some Andrei w/e. All fascinating cases for the tenured professors of psychiatry, methinks.