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.

 

[ajsdkflsdjfio]

That's the whole point. 1T PPW sucks on x86. You can't clock lunar lake any higher lol it's already clocked to the moon. There's a reason why the top bin part isn't available.


MT is fine. That's the point. But 1T isn't. Time and time again.

There's a reason why x86 battery life sux under web browsing tests (while maintaining close to plugged in perf).

IDC about all that cuz I wasn't talking about any of that, I was simply saying a 200-300% efficiency difference is impossible given Geekerwan's benchmarks and general common sense.

Also why do you keep referencing x86. It may turn out that x86 does actually have efficiency limitations compared to ARM, but you can't discern that from such basic comparisons like Lion Cove vs M4. There are a million different reasons why one core on x86 might be less efficient than another on ARM, other than the instruction set architecture differences.

 

[CouncilorIrissa]

Nope. Lunar Lake ST is pretty much equivalent to M3.

No it's not. Lunar Lake isn't efficient. It's low power.

With any profile set other than Whisper, LNL happily pulls 16-17W package power in 1t workloads.

And when set to whisper it won't reach its boost clocks.

It's how Intel achieved good battery life in reviews: when testing performance, LNL would be set to High Performance profile that allows the cores to boost. But when testing battery it'd be set to balanced profile, scoring something like this in 1t tests.
Or like this:

Spoiler: frequency pulled from .gb6

"processor_frequency": {
"frequencies": [
2128,
2230,
2510,
2745,
1811,
1546,
1414,
2412,
2439,
2490,
2144,
1981,
2102,
2631,
2599,
2640,
2641,
2722,
2627,
3021,
2982,
2873,
2547,
2668
]
},

 

[SiliconFly]

.

 

[CouncilorIrissa]

Then we should wait for tests where they test it like a M3 to get the real picture.

Wait for what?
The computerbase result contains 1t power for M3-based Air. 6W package @ 1t CBR24.

 

[techjunkie123]

400? Then 300? Now 200? Pls make up ur mind.

Cherry picked data points don't reflect real world use cases. Not even close. Like I said, LNL can easily match a M3 in both performance and power usage in most cases.


Nah... Lunar Lake doesn't do that. It works more like M3.


True.


Nope.


Nope. Some tests did show it was better than M3 in web browsing tests.


Then we should wait for tests where they test it like a M3 to get the real picture.

Where your data at? Others have provided 3 independent sources that demonstrate that lunar lake is about 300% less efficient than M3 at 1T loads.

 

[Meteor Late]

I think there is some confusion going on here.
Lunar Lake can achieve good battery life at different day to day tasks many people use with their laptop.
On the other hand, these same reviews are not measuring work done in battery mode.
The data I've seen show that Lunar Lake has good battery life, especially for an x86 chip, however, these reviews don't show if the performance is maintained while doing so, and if not, how much it degrades.

It is evident when looking at the pertinent data that Apple Silicon is a galaxy apart in terms of single core efficiency, like it's not even close. I don't understand why we are even debating this when the data is there to see. You can argue about x86 scaling better in the lower part of the curve (I am not finding data about this on Apple Silicon chips) or something like that, but these Apple chips just perform better at single core while consuming significantly less power while doing so.

 

[Meteor Late]

No it's not. Lunar Lake isn't efficient. It's low power.

With any profile set other than Whisper, LNL happily pulls 16-17W package power in 1t workloads.
View attachment 113861
And when set to whisper it won't reach its boost clocks.

It's how Intel achieved good battery life in reviews: when testing performance, LNL would be set to High Performance profile that allows the cores to boost. But when testing battery it'd be set to balanced profile, scoring something like this in 1t tests.
Or like this:
View attachment 113863

Spoiler: frequency pulled from .gb6

"processor_frequency": {
"frequencies": [
2128,
2230,
2510,
2745,
1811,
1546,
1414,
2412,
2439,
2490,
2144,
1981,
2102,
2631,
2599,
2640,
2641,
2722,
2627,
3021,
2982,
2873,
2547,
2668
]
},

That single core frequency drop is massive, out of this world. Like I said, one has to also measure work done when on battery, because it is evident that if you run at sub 3GHz you are going to inflate efficiency when on battery.

 

[Hulk]

There is no doubt that M3 is amazingly efficient and performant. In addition, the Mac OS is great for many people.

Unfortunately for me it is also completely useless, which also makes the M3 useless. I have been confounded by the Mac OS since I first encountered it in the '80's and remain so today. I am obviously too obtuse to understand it. I freely admit that. On the otherhand, I intuitively understood DOS right off the bat as well as the Windows GUI MS put on top of it. Actually my cousins had an Apple II and we did a lot of programming in Basic, that I did understand.

I thoroughly enjoy selecting each component in my x86 build, putting it together, and setting up the software in a manner that fits me perfectly. I need to know what is going on behind the curtain.

Apple has a closed infrastructure and keeps the curtain closed and secure at at times. The Wizard is in control. They control the hardware and they control the software. One of the advantages is that you can build a more efficient system when you control both sides of that interaction. If you like Apple stuff then great, go for it. But if you like the freedom, control, and massive software library of x86, well, then there you go.

Put another way you can't convince a dog person that they should give up their dog and get a cat or vice-versa. Dogs and cats. Apple and x86. Both great, both do the same thing but both totally different.

 

[ajsdkflsdjfio]

Where your data at? Others have provided 3 independent sources that demonstrate that lunar lake is about 300% less efficient than M3 at 1T loads.

300% less efficient LOL. Even if Lunarlake boosts up to 17w in heavy 1T workloads and apple only boosts up to 5W, how does that make the processor 300% less efficient? That "300% less efficient" is simply the way the power profile is programmed for that specific laptop/device. It doesn't mean the chip itself is 300% less efficient. You can literally see Apple doing the same thing in geekerwan's test with the M4 chip. He has two datapoints for 1T performance with one of the datapoints using ~10W for a single core similar to the 12W you would find on lunarlake. But you conveniently ignore this to draw a better looking conclusion.

And maybe you would try and argue that the way the power profile is programmed is still relevant to discussions of chip efficiency because real-world results matters too right? But even in that case, heavy 1T workloads are an extreme edge-case scenario and don't matter at all for real-world battery life:
1. Nobody is running heavy workloads on their thin and light laptop for more than even 5-10% of usage time.
2. Even if they are running heavy workloads, regular power behavior is vastly different than 1T power behavior, you can't give each of the 4 P-cores on lunar lake 12-15W like you do in a 1T scenario since that would far exceed the 17W TDP or even the 37W maximum rated power. Not to mention the graphics, I/O or media operations, and the e-cores.
3. MOST heavy workloads are also nT workloads, which you yourself admit are much closer in "efficiency"

You could be the best piano player in the world but for some reason are unable to play the 5-10 edge keys on each side on the piano because your arms are really short. Regardless, for the other 99.99% of songs that don't use those keys, you would still be the best piano player in the world.

For reference CB2024 nT: 563 258v standard mode, 647 M3

 

[511]

No it's not. Lunar Lake isn't efficient. It's low power.

With any profile set other than Whisper, LNL happily pulls 16-17W package power in 1t workloads.
View attachment 113861
And when set to whisper it won't reach its boost clocks.

It's how Intel achieved good battery life in reviews: when testing performance, LNL would be set to High Performance profile that allows the cores to boost. But when testing battery it'd be set to balanced profile, scoring something like this in 1t tests.
Or like this:
View attachment 113863

Spoiler: frequency pulled from .gb6

"processor_frequency": {
"frequencies": [
2128,
2230,
2510,
2745,
1811,
1546,
1414,
2412,
2439,
2490,
2144,
1981,
2102,
2631,
2599,
2640,
2641,
2722,
2627,
3021,
2982,
2873,
2547,
2668
]
},

You do realsiy it is a bug in windows power plan which is causing this ona ASUS laptop? cause it is not even boosting to its peak

 

[techjunkie123]

300% less efficient LOL. Even if Lunarlake boosts up to 17w in heavy 1T workloads and apple only boosts up to 5W, how does that make the processor 300% less efficient? That "300% less efficient" is simply the way the power profile is programmed for that specific laptop/device. It doesn't mean the chip itself is 300% less efficient. You can literally see Apple doing the same thing in geekerwan's test with the M4 chip. He has two datapoints for 1T performance with one of the datapoints using ~10W for a single core similar to the 12W you would find on lunarlake. But you conveniently ignore this to draw a better looking conclusion.

And maybe you would try and argue that the way the power profile is programmed is still relevant to discussions of chip efficiency because real-world results matters too right? But even in that case, heavy 1T workloads are an extreme edge-case scenario and don't matter at all for real-world battery life:
1. Nobody is running heavy workloads on their thin and light laptop for more than even 5-10% of usage time.
2. Even if they are running heavy workloads, regular power behavior is vastly different than 1T power behavior, you can't give each of the 4 P-cores on lunar lake 12-15W like you do in a 1T scenario since that would far exceed the 17W TDP or even the 37W maximum rated power. Not to mention the graphics, I/O or media operations, and the e-cores.
3. MOST heavy workloads are also nT workloads, which you yourself admit are much closer in "efficiency"

You could be the best piano player in the world but for some reason are unable to play the 5-10 edge keys on each side on the piano because your arms are really short. Regardless, for the other 99.99% of songs that don't use those keys, you would still be the best piano player in the world.

For reference CB2024 nT: 563 258v standard mode, 647 M3
View attachment 113878

1T not nT. M4 beats lunar lake by 40%, at 10 W, vs lunar lake at 17 W. That's still a big gap lulz. Saying you can pick a more efficient point on the curve is ofc true but is completely irrelevant. You're comparing two thin and light laptops lmao.

My point is simple: macs last way longer than x86 windows laptop on most light-medium loads (barring just video playback), which is how most people use their laptops. This is just a simple reality. I personally don't like this reality, but it's seriously making me consider switching to a mac. AMD/Intel need to figure this out. Qualcomm already has with oryon.

 

[poke01]

There is no doubt that M3 is amazingly efficient and performant. In addition, the Mac OS is great for many people.

Unfortunately for me it is also completely useless, which also makes the M3 useless. I have been confounded by the Mac OS since I first encountered it in the '80's and remain so today. I am obviously too obtuse to understand it. I freely admit that. On the otherhand, I intuitively understood DOS right off the bat as well as the Windows GUI MS put on top of it. Actually my cousins had an Apple II and we did a lot of programming in Basic, that I did understand.

I thoroughly enjoy selecting each component in my x86 build, putting it together, and setting up the software in a manner that fits me perfectly. I need to know what is going on behind the curtain.

Apple has a closed infrastructure and keeps the curtain closed and secure at at times. The Wizard is in control. They control the hardware and they control the software. One of the advantages is that you can build a more efficient system when you control both sides of that interaction. If you like Apple stuff then great, go for it. But if you like the freedom, control, and massive software library of x86, well, then there you go.

Put another way you can't convince a dog person that they should give up their dog and get a cat or vice-versa. Dogs and cats. Apple and x86. Both great, both do the same thing but both totally different.

Agree. This is why I choose the best of both worlds. My desktop will always be a x86 PC and my laptop from Apple.

 

[poke01]

AMD/Intel need to figure this out.

Intel can do it, they can just need better P core arch. The sad thing is Lunar Lake style SoC is the last of its kind for a long while from Intel.

 

[ajsdkflsdjfio]

Intel can do it, they can just need better P core arch. The sad thing is Lunar Lake style SoC is the last of its kind for a long while from Intel.

I think Panther-lake is pretty similar to Lunar-Lake SoC design with the difference being that the graphics are in a separate tile, but that is not too big of a deal with how distinct the GPU and CPU normally are anyways. It's not like Arrow-lake with a separate SOC tile causing a bunch of performance issues.

As for P-cores yea they are plain uncompetitive for anything other than max performance, I don't think Intel will be competitive perf/area or power until unified core or some other massive overhaul away from the current p-core architecture.

 

[poke01]

For thin 'n lights, 1T is more important. Blender and cinebench only provide a technical comparison not a real-world one.

I would say this Pugetbench/Speedometer chart is more important than any Cinebench test for these SoCs.





Intel made amazing progress from MTL to LNL. Apple still has a lead with M3 despite using the same node as Lunar Lake.

 

[poke01]

I think Panther-lake is pretty similar to Lunar-Lake SoC design with the difference being that the graphics are in a separate tile, but that is not too big of a deal with how distinct the GPU and CPU normally are anyways. It's not like Arrow-lake with a separate SOC tile causing a bunch of performance issues.

Panther Lake is different from Lunar Lake. For starters it doesn't use PMICs anymore but switches back to VRMs.

Its a more cost efficient design for the mass market OEMS

 

[ajsdkflsdjfio]

For thin 'n lights, 1T is more important. Blender and cinebench only provide a technical comparison not a real-world one.

I would say this Pugetbench/Speedometer chart is more important than any Cinebench test for these SoCs.

Why is 1T more important in thin and light's? I'd argue that it's less important since thin and lights are rarely pushed to do extremely heavy tasks which are limited by 1T performance. Also aren't Pugetbench/Speedometer nT tests?

Also I agree, technical comparisons are not as useful for real-world use but in real-world use the consensus is that lunar-lake provides competitive battery life even to snapdragon/apple. It's only in the technical benchmarks where lunar-lake efficiency problems are shown.

PugetBench and especially speedometer are not at all useful for real-world scenarios if even considered to be accurate. Like somehow the speedometer test has a meteor-lake CPU beating a zen5 monolithic laptop chip at similar TDP? There's clearly something wrong with that test and/or whatever workload they use is useless for any comparisons theoretical or real-world.

Panther Lake is different from Lunar Lake. For starters it doesn't use PMICs anymore but switches back to VRMs.

Its a more cost efficient design for the mass market OEMS

Yes Panther-Lake is not as finely tuned as Lunar Lake for low power, but at the same time it has similarities which make it a more effective mobile chip than arrow-lake. In addition to the fact that it lacks a separate SOC tile, it also has LP-E cores in the compute tile like lunar lake does,. which will massively help with battery life compared to arrow-lake and meteor-lake.

 

[OneEng2]

Something other than geekbench 6 like a real review would be a better judge of the two.

Totally agree.

ARM still have the efficiency crown.

Well, I am not going to dispute that at all; however, I would say that since ARM has been being optimized for efficiency for decades now while x86 has been optimized for maximum performance, and ARM originated in an environment that until the last 5 years could have cared less about MT while x86 has been taking on highly threaded and shared processing for decades.

It is not unreasonable to expect one to be more efficient in single thread loads while the other performs best for the market it was designed for.

The fact of the matter is, there is no "secrete sauce" in either x86 or ARM. x86 decoder must be a little more complex in order to handle variable length instruction and data (first turning them into fixed length instruction and data); however, aside from that they both utilize the exact same kinds of logic blocks to accomplish what they do.

IMO, it is silly to suggest that x86 can't reach the efficiency of ARM or that ARM can reach the performance of x86. It is just a matter of design priorities IMO.

I would like to know where you got the 98% number for x86 in corporate. Even in enterprise I don’t believe it is that high. IBM Z systems have continued to maintain a presence. In cloud Amazon is on their fourth Graviton processor and may be using them for a lot of internal load.

Yea, me too. I can't find it again. The best number I can find now is just the overall laptop number of 94.3% x86, but that is consumer as well.

The main advantage to ARM is the cost of IP if you can cover the NRE with adequate volume. AMD is still making money on x86 and they have about a 40% margin over fab margin. Anyone of the cloud providers with significant volume can potentially reduce costs by fabbing their own processor. ARM server design volume is still relatively small but V2 volumes are picking up which spreads design costs.

If you are google or apple, and have your own OS, then ARM is the only game in town for all the reasons you suggest. Additionally, you can customize your own processor to work best with your OS.

Servers are another animal completely. The processor must run standard OS's and off-the-shelf apps (and a lot of them) well. Here, having a standard processor that development teams can design for on standard operating systems is critical. Making your own processor, compiler, operating system, and applications can only be done by the mighty few.

 

[511]

Also M3 ST is superior to lunar Lake due to their better architecture and vertical sw integration

 

[511]

Panther Lake is different from Lunar Lake. For starters it doesn't use PMICs anymore but switches back to VRMs.

Its a more cost efficient design for the mass market OEMS

It has optional PMIC iirc many OEMs are not happy with PMICs same with snapdragon cause it uses 6 PMIC and upto 8! No OEM like this Apple can get away cause they are Apple

 

[Hulk]

Why is 1T more important in thin and light's? I'd argue that it's less important since thin and lights are rarely pushed to do extremely heavy tasks which are limited by 1T performance. Also aren't Pugetbench/Speedometer nT tests?

Also I agree, technical comparisons are not as useful for real-world use but in real-world use the consensus is that lunar-lake provides competitive battery life even to snapdragon/apple. It's only in the technical benchmarks where lunar-lake efficiency problems are shown.

PugetBench and especially speedometer are not at all useful for real-world scenarios if even considered to be accurate. Like somehow the speedometer test has a meteor-lake CPU beating a zen5 monolithic laptop chip at similar TDP? There's clearly something wrong with that test and/or whatever workload they use is useless for any comparisons theoretical or real-world.

Yes Panther-Lake is not as finely tuned as Lunar Lake for low power, but at the same time it has similarities which make it a more effective mobile chip than arrow-lake. In addition to the fact that it lacks a separate SOC tile, it also has LP-E cores in the compute tile like lunar lake does,. which will massively help with battery life compared to arrow-lake and meteor-lake.

I see your point but I tend to agree with Poke01 regarding ST being very important for thin and light. But there is some nuance here.

By ST I believe Poke01 is referring to the majority of applications that are only using a handful or less of threads. Office applications, web browsing, PS lighter editing work, vector illustration, and other day-to-day go-to applications. A couple performant cores are necessary in order for the thin and light to feel "snappy." I know when I have heavy lifting to do I try to wait to do it on my desktop. For example, I can edit video on my laptop but I'm not going to render on it, or I'll do the lighter editing work, but I'm not going to apply plugins that are compute intensive like Mercalli stabilization, or do detailed keyframing on a small monitor. I can get the basics down on the laptop and then finish the heavy lifting details and rendering on the desktop.

So in short, for most applications that don't require a massive number of cores and threads a thin and light will "feel" more like a desktop if it has good ST performance.

There is no way I'm going to punish myself on a laptop with photo/video AI, heavy video work, or other compute intensive work that also benefits from a large display when I have a 9950X and a 27" 4k monitor at home.

But as always, everybody's work flow is different and we tend to think of our individual use case as "most people."

 

[poke01]

Like somehow the speedometer test has a meteor-lake CPU beating a zen5 monolithic laptop chip at similar TDP?

The Zen5 mobile chip is likely not hitting its 5.1GHz peak and its not optimised for ST but MT. Zen5 desktop is much much better, there it scores from 40-45 in Speedometer 3.

MTL is really bad, if I had to choose between MTL and mobile Zen5 I would still pick AMD.

It has optional PMIC iirc many OEMs are not happy with PMICs same with snapdragon cause it uses 6 PMIC and upto 8! No OEM like this Apple can get away cause they are Apple

Its expensive for sure.

Also aren't Pugetbench/Speedometer nT tests?

I consider them to be 1t tests or at best lightly threaded tests, as they don't fully saturate all the cores like Cinebench and Blender do.

nT tests like Blender like to use all cores and SMT. This is where AMD shines. But you can see with the 258V 4P4E/8T and M3 4P4E/8T its close.

The M3 15" Air has no active cooling, uses slower memory and uses 20 watts but still wins against the 258V.

 

[MoistOintment]

The whole "death of x86" argument is really missing the forest for the trees. Noone is seriously arguing that x86 will cease to exist anywhere in the near future. x86 will live on in some form or another for many decades to come.

The death of x86 is more about the x86 TAM shrinking to the point that it's much less viable (or non-viable) as a mass market consumer product.

Someone else in this thread said ARM was "only" 10% of the client laptop market....as if ARM going from 0% to 10% marketshare in just 4 years in one of x86's last (of 2) markets isn't fairly remarkable growth.

What happens when ARM is 50% of the laptop market? Would Intel and AMD fighting over that remaining 50% even make financial sense?

And someone else asked what benefit would one personally get from switching to ARM - for most, no direct benefit or downside. Besides the benefit of the client laptop market not being beholden to a CPU duopoly.

Point being, the "death of x86" is not talking about 0% marketshare. It's the point in which the x86 TAM shrinks to much, that the transition to ARM accelerates beyond the point of no return and where the x86 TAM is too small to support the scale necessary to justify its continued development.

 

[Thunder 57]

The whole "death of x86" argument is really missing the forest for the trees. Noone is seriously arguing that x86 will cease to exist anywhere in the near future. x86 will live on in some form or another for many decades to come.

The death of x86 is more about the x86 TAM shrinking to the point that it's much less viable (or non-viable) as a mass market consumer product.

Someone else in this thread said ARM was "only" 10% of the client laptop market....as if ARM going from 0% to 10% marketshare in just 4 years in one of x86's last (of 2) markets isn't fairly remarkable growth.

What happens when ARM is 50% of the laptop market? Would Intel and AMD fighting over that remaining 50% even make financial sense?

And someone else asked what benefit would one personally get from switching to ARM - for most, no direct benefit or downside. Besides the benefit of the client laptop market not being beholden to a CPU duopoly.

Point being, the "death of x86" is not talking about 0% marketshare. It's the point in which the x86 TAM shrinks to much, that the transition to ARM accelerates beyond the point of no return and where the x86 TAM is too small to support the scale necessary to justify its continued development.

Were you around in the early 90's? Sure things have changed, but I've been hearing this forever. And at this point, it is probably worth talking about in a new thread

 

[ajsdkflsdjfio]

I see your point but I tent to agree with Poke01 regarding ST being very important for thin and light. But there is some nuance here.

By ST I believe Poke01 is referring to the majority of applications that are only using a handful or less of threads. Office applications, web browsing, PS lighter editing work, vector illustration, and other day-to-day go-to applications. A couple performant cores are necessary in order for the thin and light to feel "snappy." I know when I have heavy lifting to do I try to wait to do it on my desktop. For example, I can edit video on my laptop but I'm not going to render on it, or I'll do the lighter editing work, but I'm not going to apply plugins that are compute intensive like Mercalli stabilization, or do detailed keyframing on a small monitor. I can get the basics down on the laptop and then finish the heavy lifting details and rendering on the desktop.

So in short, for most applications that don't require a massive number of cores and threads a thin and light will "feel" more like a desktop if it has good ST performance.

There is no way I'm going to punish myself on a laptop with photo/video AI, heavy video work, or other compute intensive work that also benefits from a large display when I have a 9950X and a 27" 4k monitor at home.

But as always, everybody's work flow is different and we tend to think of our individual use case as "most people."

I agree, I was just saying that although 1T performance is always going to be important even in thin & light, but even so a 50% faster/more efficient core isn't going to make your experience 50% snappier or more responsive. As long as your processor isn't super old and you have enough cores/ram to support your applications, the snappiness you feel is largely dependent on operating system and the programs you use. Therefore 1T performance is less important than in desktop, although still very important.

Although I would still love my laptop to have excellent 1T performance/efficiency, I would still argue that the 50% 1T efficiency in benchmarks translates less into real-world use than one might imagine. My M1 macbook air still serves me well enough for light work. Buying a 50%+ faster M4 laptop would do me no good for what I typically use a laptop for.

The Zen5 mobile chip is likely not hitting its 5.1GHz peak and its not optimised for ST but MT. Zen5 desktop is much much better, there it scores from 40-45 in Speedometer 3.

MTL is really bad, if I had to choose between MTL and mobile Zen5 I would still pick AMD.

I understand that, my point was that speedometer having a better result for meteor lake compared to Strix Point doesn't really give much creedence to the accuracy or usefulness of that benchmarking program. Zen5 mobile was obviously nerfed vs desktop, but so is meteor lake which is a trash and older product regardless. The HX 370 has around a 20-30% 1T performance lead over the 185H and is monolithic to boot.

I consider them to be 1t tests or at best lightly threaded tests, as they don't fully saturate all the cores like Cinebench and Blender do.

nT tests like Blender like to use all cores and SMT. This is where AMD shines. But you can see with the 258V 4P4E/8T and M3 4P4E/8T its close.

The M3 15" Air has no active cooling, uses slower memory and uses 20 watts but still wins against the 258V.

Meh, nT doesn't mean 100% all cores workloads, it just means workloads which use more than 1 core and puget bench especially benefits from having more cores. Also the 15 inch Macbook air chassis with no fans isn't likely to limit the performance of the M3 chip by a large margin in a workload running only a few minutes at most. Even on the 13 inch laptop, performance only starts to throttle after a couple minutes of a heavy workload and results in a gradual drop in performance over time rather than an instance kneecapping of performance.

Anyways I agreed with your point that "real-world" usage(if you can define it well enough) is a better indicator of how good a laptop is for most consumers than technical tests. On the surface pugetbench and speedometer may seem like they give a good idea of "real-world" usage performance since one tests photoshop and another web-browsing, but at the end of the day they are a far cry from the workflow of any usual person and are just as bad as cinebench and blender for determining a CPU's actual effect on users work/play.

But this is irrelevant to the fact that these benchmarks themselves are flawed for a CPU to CPU comparison. Pugetbench is also heavily reliant upon the GPU and therefore will have it's scores differ significantly even on the same CPU using different GPUs. With the M3 chip having around a 20% stronger GPU, the Puget bench scores you referenced aren't reliable.

Speedometer also seems to be very finicky even varying by +-10% based on the browser you are using. In the first place it seems to heavily favor apple even in scenarios where their CPUs are actually weaker than their competitors. For example I scored a 29.4 on speedometer 3.0 on my M1 macbook air with adblock, Ublock origin, and lastpass extensions installed. Compare this to the ~30 or below scores that many desktop raptor-lake and zen4 owners report and you see that there is clearly something different about MacOS that causes speedometer to perform better. Run to run variance is also pretty bad too with a up to 10% difference between different runs on my M1 macbook and PC without changing anything and spacing the tests out by 1 min, no other benchmark has this much run to run variance.

Also despite scoring around a 10-12.5(depending on the run) on my 9600k desktop PC, I vastly prefer the browsing experience on my desktop PC vs my M1 macbook air that scores close to a 30 on speedometer. I find that my M1 macbook slows down in significantly more scenarios while multitasking/browsing the web even with the same amount of ram and being plugged in.

This was around the typical score I found for zen4 AMD users and it's still on par/weaker than my M1 Macbook air which has around a 10-20% weaker 1T performance and much slower ram among other things.