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.

 

[jdubs03]

I think you have mistaken me, Intel uses M3 instead of M4 because the latter is not in laptops is what I meant.

Although the definition of ultra slim systems is up to interpretation, I guess Intel means Windows or macOS machines here.

Huh? He was taking about the Apple M4 not being compared to LNL on the slides because it’s behind (I could also argue it’s not x86 and the comparison could still be valid but I digress).

I cannot find post-launch SPEC2017 scores for SD X Elite, bar this one which was before launch.

That was running at 3.95 GHz. The SKU Intel compares with is running at 3.4 GHz. So it should score roughly 7.05 at 3.4 GHz instead of 8.19. So with the Intel Compiler, +61% over the SD X Elite puts the 288V at 11.35

That is basically tied with M4 in the iPad Pro, as per Geekerwan.

Obvious caveat is that this is ICC on Intel vs who-knows-what on the SD X Elite.

In the image for geekerwan it explicitly shows the highest end MTL, and the HX 370 compared to both the M3 and M4. Based on the percentage difference in Intel’s own slides the int score calculates out to around 8.1 per that image; it cannot be anywhere near the M4. Even that is 29% higher than the 185H.

Now here is the hot chips image for the Snapdragon reference device (I presume the 80w design and not the 23w, someone correct me if I’m wrong there):

Definitely has to be a compiler difference for this (and a fully cooled reference design). Because when considering this data, the percentage differences make less sense.

They certainly don’t like to make it easy to do 1on1 comparisons but my point is there is no way based on the data that LNL 1T has approached M4, let alone M3. Just look at the Geekbench scores.

 

[DokiDoki]

Huh? He was taking about the Apple M4 not being compared to LNL on the slides because it’s behind (I could also argue it’s not x86 and the comparison could still be valid but I digress).


In the image for geekerwan it explicitly shows the highest end MTL, and the HX 370 compared to both the M3 and M4. Based on the percentage difference in Intel’s own slides the int score calculates out to around 8.1 per that image; it cannot be anywhere near the M4. Even that is 29% higher than the 185H.

Now here is the hot chips image for the Snapdragon reference device (I presume the 80w design and not the 23w, someone correct me if I’m wrong there):
View attachment 106774
Definitely has to be a compiler difference for this (and a fully cooled reference design). Because when considering this data, the percentage differences make less sense.

They certainly don’t like to make it easy to do 1on1 comparisons but my point is there is no way based on the data that LNL 1T has approached M4, let alone M3. Just look at the Geekbench scores.

What is the fmax of X Elite? 4.3 GHz? So assuming this Hot Chips slide is showing measurements at that frequency, the SKU that Intel compared against, which runs at 3.4 GHz, should score 10.64*3.4/4.3 = 8.41 under Linux.

So, going by Intel's numbers the 288V is then scoring in excess of 13pts in SPECint2017 rate, which is firmly M4 territory. In fact, it is higher. But we'll see when M4 launches in the Macbooks. I think it will be close.

 

[vanplayer]

Real bench begins to float around

the windows scheduler is a mess that it cannot recognize big little correctly

https://twitter.com/x/status/1831140640094347627

 

[DokiDoki]

Real bench begins to float around

the windows scheduler is a mess that it cannot recognize big little correctly

https://twitter.com/x/status/1831140640094347627

Calling CPU-Z benchmark "real bench" is a travesty.

 

[jdubs03]

What is the fmax of X Elite? 4.3 GHz? So assuming this Hot Chips slide is showing measurements at that frequency, the SKU that Intel compared against, which runs at 3.4 GHz, should score 10.64*3.4/4.3 = 8.41 under Linux.

So, going by Intel's numbers the 288V is then scoring in excess of 13pts in SPECint2017 rate, which is firmly M4 territory. In fact, it is higher. But we'll see when M4 launches in the Macbooks. I think it will be close.

The X80 boosts to 4.2GHz. The X84 and reference goes to 4.3Ghz. The slide showing the comparison says X80 (whereas the claim and statement slide says X84 which isn’t great because it confuses things).

But I’m just going by the comparison of LNL 288V vs HX 370, and then applying that ratio ~1.25 to the spec2017int(1-copy) score for Geekerwan to the 185H. That results in 8.1.

Now compared to David Huang’s testing at https://blog.hjc.im/spec-cpu-2017
the HX 370 scores 9.72, applying the same ratio gets to 12.15. The M3 Pro scores 11.8, unfortunately there is no standard M3 score (but it performs just as good or better in GB and CB 1T so the assumption is it’s similar, I’ll add the caveat that the M2 Pro does about 5% better than the M2, but also different architectures to the M3 and M4). When applying the differential of M3 to M4 from geekerwan there is an 8% improvement, so the M4 Pro would be about 12.77. I’d imagine the M4 is similar, but could be less.

Point is the 288V is going to score lower than the M4-series. It doesn’t just magically leap so far ahead of its predecessor with a 14% IPC bump and node shrink. All the single-thread tests will bare this out.

Plus, if it did score similar to the M4, wouldn’t you think they would be broadcasting that all over the place?

 

[vanplayer]

Calling CPU-Z benchmark "real bench" is a travesty.

It's just because there was an old fake pic that showing 285k CPU-Z 1T more than 1100pts few weaks ago and people in this forum seem to believed it, that's why I call it REAL this time. I didn't comment on whether a bench itself is good or not.

Anyway 7-9% ST uplift from RPL fits the expectation pretty well.

 

[poke01]

Dell Lunar buy page

Dell has their buy page up, @Hulk you buying?

 

[cebri1]

The results are pretty much aligned with expectations. 265K should have around the same performance of the 13/14900K.

 

[Henry swagger]

https://wccftech.com/intel-core-ult...cpu-pictured-5-percent-faster-benchmark-leak/ overclocking should bring even bigger score ☺

 

[igor_kavinski]

Dell Lunar buy page

Dell has their buy page up, @Hulk you buying?

Looks too expensive for only 16GB RAM.

 

[poke01]

Looks too expensive for only 16GB RAM.

N3B is costly plus look at whole package definitely more worth it than the X Elite one and the price reflects that.

16GB is fine for most people. The most disappointing thing about the new Dell is the lack of ports. The HP ultra book is nice but that’s $1499 for the base Ultra 5 whereas this dell gets an ultra 7 for $1399.

 

[naukkis]

Neither is the M4 😉

Based on this:
View attachment 106769
and this:
View attachment 106770
Performance in spec2017 should thus be somewhere around 8.1. So the M4 is around 32% faster in integer for 1T in the iPad Pro. For the MacBook, one would think it would score a little better.

What is impressive though is what looks like a 29% jump from the 9 185H. Strange though as that’s way more than the 14% IPC increase for LNC vs RWC? Am I missing something here?

Yup, that slide is comparing IPC performance at fixed 4ghz. You cannot compare 1T performance based on that alone without freq.

 

[SiliconFly]

.

 

[cebri1]

Can we please remove "Meteor Lake" from the thread name since it's kinda passé and Lunar Lake time has come?

Isn’t the forum going to die?

 

[jdubs03]

Yup, that slide is comparing IPC performance at fixed 4ghz. You cannot compare 1T performance based on that alone without freq.

This is fair. Good catch.
The best I got from the Geekerwan slides are this for the HX 370:

Which would put the 1Tint score for the 288V at 10.0.

Unfortunately the best I got for full speed M4 is this (which was cooled with LN2, there was another slide that had M4 listed both values were the same as the iso-4.0Ghz slide):

With active cooling (which is a fair comparison as I think its competitors have fans), I doubt the performance is reduced significantly.

Time will tell how close they are. One thing I'm kind of surprised about is no X Elite (preferably the X84) being tested in spec2017 outside of that one first-party image I posted.

 

[jdubs03]

Isn’t the forum going to die?

Well, we’re still trying to figure out where we should all go if it does. That is TBD.

 

[SiliconFly]

.

 

[DavidC1]

16GB is fine for most people. The most disappointing thing about the new Dell is the lack of ports. The HP ultra book is nice but that’s $1499 for the base Ultra 5 whereas this dell gets an ultra 7 for $1399.

Two USB-C ports which will sometimes be taken up by charging, no audio jack ports either.

HP looks better on the port selection plus it's lighter. I'd take it over Dell. The XPS 12 I got had a widely known screen issue so I couldn't even sell nor replace the screen.

 

[Det0x]

Dunno if true, but this sounds pretty snake oil ™ to me (?)


Also, are they trying to tell us strix point have ~117ns latency to memory ?

btw the Meteor Lake interconnet fabric is what Arrow Lake resuses right ?
Will AR have ~126ns latency to memory ? Having a hard time believing that..

But looking at these results together with the recent stock collapse, maybe the prayers from the CEO is warranted 🤣

 

[coercitiv]

16GB is fine for most people.

with the new "AI" features 16GB is the bare minimum.

Spoiler: Powerful AI

 

[poke01]

with the new "AI" features 16GB is the bare minimum.

Spoiler: Powerful AI

View attachment 106799

thats marketing speak, 16GB ought be enough for these:

This is whats coming in November.

• Live Captions with translation
• Windows Studio Effects (Background effects: standard blur, portrait blur, eye contact, and automatic framing)
• Cocreator in Paint
• Photos: Restyle image and Image Creator

Now Recall may need 32GB RAM for the best experience.

Copilot+ PCs expand availability with new AMD and Intel silicon

In May, we announced a new category of devices: Copilot+ PCs. These PCs are the fastest, most intelligent Windows PCs ever built, featuring a new system architecture d

blogs.windows.com

 

[poke01]

Dunno if true, but this sounds pretty snake oil ™ to me (?)

Yeah its true. https://browser.geekbench.com/ai/v1/43936
But thats to be expected as AMD has more cores and AVX512 but Intel felt the need to win.

 

[DokiDoki]

Also, are they trying to tell us strix point have ~117ns latency to memory ?

That number is from Chips & Cheese. They measure ~130ns for Strix Point.

They do seem to have goofed up OpenVINO vs ONNX though.

 

[igor_kavinski]

16GB is fine for most people.

Maybe but SD Elite X 32GB laptop I can get delivered tomorrow for $1200. Will be forced to wait for LNL to get a bit stale so I can afford it. It's still a better bet with 32GB RAM due to guaranteed software support (well, if Intel survives that is or they don't EOL LNL too soon).

What I find annoying about AMD is that they have not been able to make their NPU easily accessible so far. They have like what? 3rd gen NPU in Ryzen AI now? What can I do with it if I get Ryzen AI laptop? Nothing I can think of, other than generating images with their AI suite or going really deep into the AI rabbit hole with their Versal AI libraries (not gonna do that. That's stupid).

 

[Det0x]

That number is from Chips & Cheese. They measure ~130ns for Strix Point.

They do seem to have goofed up OpenVINO vs ONNX though.

Yeah seems like you are correct 👍


But its really strange to me that these mobile chips have so high latency compared to desktop, this is one of my old runs on regular Zen5
(on latest tune i haven't made made public yet, i'm sub 74ms)


This question still remains, will be interesting when we get closer to launch

btw the Meteor Lake interconnet fabric is what Arrow Lake resuses right ?
Will AR have ~126ns latency to memory ?