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.

 

[ondma]

I think arrow lake refresh will use n3E which is better than n3B

I just dont get intel. How can they be so inept. Not only can they not produce enough of their own nodes to satisfy their own demand, but when they go to TSMC they get an inferior node from them.

 

[Racan]

Thoughts on this? Seems quite ambitious. GLC IPC based on GB6 is 511 pts/GHz, so 2.25x is 1150 pts/GHz. 3x is even crazier if that’s 2030 timeframe.
M4 is 833 pts/GHz. To get to around that 1150 would need an 8.5% increase per year.

https://www.reddit.com/r/hardware/comments/1f0j0kh/comment/ljt2bqf

View attachment 106146

I hope AMD scoops them up.

 

[DavidC1]

I still wonder what is their main target group for these...

The point is battery life superiority in an x86 platform so it has no compatibility issues and GPU that's way better than on Snapdragon.

All having 4+4 means it doesn't matter as much if you get Ultra 5.

Thoughts on this? Seems quite ambitious. GLC IPC based on GB6 is 511

View attachment 106146

There was another comment saying when he left the Royal Core group the performance on SpecCPU wasn't that impressive compared to the other teams projects at least for the first version.

I would just say it was way too ambitious and probably unrealistic. Computers already advance at a breakneck pace. You can't expect the increase in pace of that to increase, can you?

 

[DavidC1]

Need both core sizes, clock ceilings, perf/watt and clock comparisons, but as it is based on rough data it would appear you'd need around 1.4-1.5 more Skymont cores than Zen 5 cores (with Zen 5 operating at similar frequencies ) , and around 1.2x Zen 5c (both at max clock speeds) to match throughput

Zen 5c is slower than regular Zen 5 by few %. The latter is 6% faster in Int. Regular Zen 5 is 10% faster than Zen 4 and if Skymont is 2% faster than RPL then the gap is essentially zero.

Since Skymont in Arrowlake clocks higher than Gracemont despite the P core sister losing frequency, I would wager on Skymont having significant clock advantage over 5C, if power is damned. It is possible if the 1.5mm+ size is true then maybe 30% extra core size is taken up to increase clock potential(which I think is stupid).

We have another example of very different approaches reaching a similar end result.

 

[Abwx]

Zen 5c is slower than regular Zen 5 by few %. The latter is 6% faster in Int. Regular Zen 5 is 10% faster than Zen 4 and if Skymont is 2% faster than RPL then the gap is essentially zero.

SKT has 4% better perf/clock than Zen 3 in CB R23 and likely 5% slower in CB R15, for R23 that s almost s 100% confirmed.

 

[Henry swagger]

https://wccftech.com/intel-core-ultra-7-268v-lunar-lake-even-better-new-cpu-gpu-benchmarks/ 17w matching 120watt amd 7900x .. lumar lake is intel's m1 moment 😀🥇 .. apple changed the industry with the m1 in 2020 🥇

 

[jdubs03]

There was another comment saying when he left the Royal Core group the performance on SpecCPU wasn't that impressive compared to the other teams projects at least for the first version.

I would just say it was way too ambitious and probably unrealistic. Computers already advance at a breakneck pace. You can't expect the increase in pace of that to increase, can you?

Got a quick link there? I was searching the comment thread but didn’t spot it.

 

[511]

Thoughts on this? Seems quite ambitious. GLC IPC based on GB6 is 511 pts/GHz, so 2.25x is 1150 pts/GHz. 3x is even crazier if that’s 2030 timeframe.
M4 is 833 pts/GHz. To get to around that 1150 would need an 8.5% increase per year.

https://www.reddit.com/r/hardware/comments/1f0j0kh/comment/ljt2bqf

View attachment 106146

You called my name 511 is here 😉

 

[511]

SKT has 4% better perf/clock than Zen 3 in CB R23 and likely 5% slower in CB R15, for R23 that s almost s 100% confirmed.

This is reference to SKT in LNL or on Ring Bus?

 

[cebri1]

I need a new laptop in a bad way. My Surface Laptop 2 has served me well and is still serving me but I'm out of SSD space, trackpad went nuts about a year ago (disabled it), and somehow I cracked the screen. Plus the Skylake quad core is completely overmatched for even light video editing at 4K, Photo/Video AI, and many other apps.

I'm most likely going to buy a LL laptop shortly after launch assuming there are some decent choices. I'll bench the heck out of it and let you all know how it does.

One wonders if a 16 core chadmont would have a market for a “low power” video/photo editing machine.

Nope cause intel 4 meteor lake has IDM Margins as well with lunar lake TSMC there is no margin for manufacturing besides packing

That’s not how it works: gross margin = average price per unit - average cost per unit. It doesn’t matter if the product was built in house or it was outsourced. Intel 4 is quite expensive per Intel’s last earning call, maybe it was cheaper to outsource it to TSMC.

 

[SiliconFly]

.

 

[poke01]

https://twitter.com/x/status/1827623565593825733

Some improvement over last time for the 288V, 5.1GHz SKU

 

[jdubs03]

https://twitter.com/x/status/1827623565593825733

Some improvement over last time for the 288V, 5.1GHz SKU

Definitely has some inconsistencies. Like Cinebench nT decreasing. And it’s odd that ST wasn't measured. But ST does look on-par/slightly ahead of its direct competitor.

 

[Abwx]

This is reference to SKT in LNL or on Ring Bus?

For the 285K.

https://twitter.com/x/status/1827623565593825733

Some improvement over last time for the 288V, 5.1GHz SKU

Previously It was 10212 pts in CB R23@30W, wich is about the same score at same power than a 18 months old 6C/12T 7540U, so they are quite late to the party, dunno why the excitement.

 

[Henry swagger]

Just went through the link. It is quite amazing that a 17w lunar lake part can actually match a 120w 7900X in ST. Lunar lake products are gonna be pretty exciting assuming they’re priced right.

Competition has only a very small market share in laptops. At this level of performance, Lunar Lake might actually wipe it totally out of existence.

Yeah intel will increase market share even more if tsmc can supply 10 or 15 millions chips

 

[511]

That’s not how it works: gross margin = average price per unit - average cost per unit. It doesn’t matter if the product was built in house or it was outsourced. Intel 4 is quite expensive per Intel’s last earning call, maybe it was cheaper to outsource it to TSMC.

Yeah but they said they ran hot lot for Meteor lake demand outstriping their estimation

 

[mikk]

The main excitement comes from the much increased efficiency/battery life in many real world usage scenarios like video playback or browsing or any other light workloads which is unheard-of for x86. Sure with 8 threads it won't set new efficiency records under full load but who does this constantly on a (low power) mobile device. If someone needs to do rendering tasks regularly he should better buy something with more cores and threads.

 

[511]

For the 285K.



Previously It was 10212 pts in CB R23@30W, wich is about the same score at same power than a 18 months old 6C/12T 7540U, so they are quite late to the party, dunno why the excitement.

We for extra 13W 76% more power we got 19% more performance so LNL doesn't scale well at 30W in terms of CPU power 17W is the best point to operate at

 

[511]

Yeah intel will increase market share even more if tsmc can supply 10 or 15 millions chips

TSMC can supply that considering Intel's Pre Pay and volumes which are larger than AMD and Qualcomm combined don't forget they supply Apple so volume is never a TSMC Problem

 

[Abwx]

We for extra 13W 76% more power we got 19% more performance so LNL doesn't scale well at 30W in terms of CPU power 17W is the best point to operate at

That s 15W CPU + 2W RAM and 28W CPU + 2W RAM, so the CPU power increase by 87% while the CB score increase by 24.5%, this amount to power increasing at a 2.85 exponent rate of frequency, wich is more than previous TSMC s processes, so it s either the process or the design that did cut some corners by increasing density at the expense of perf/watt at the higher range of the TDP, not that it make a huge difference.

 

[cebri1]

The only thing that slightly worries me about these new CPUs with already midrange iGPUs is cooling capabilities of the laptop. I currently have a G14 with Ryzen 9 with a 4060, and the laptop is much better cooled when the dGPU enabled despite drawing more power. I guess the APU gets quite hot when both the 780M and the R9 are enable.

 

[511]

That s 15W CPU + 2W RAM and 28W CPU + 2W RAM, so the CPU power increase by 87% while the CB score increase by 24.5%, this amount to power increasing at a 2.85 exponent rate of frequency, wich is more than previous TSMC s processes, so it s either the process or the design that did cut some corners by increasing density at the expense of perf/watt at the higher range of the TDP, not that it make a huge difference.

Score increased by only ~20% also it is package power which can be redistributed where is necessary

 

[Magio]

Previously It was 10212 pts in CB R23@30W, wich is about the same score at same power than a 18 months old 6C/12T 7540U, so they are quite late to the party, dunno why the excitement.

The excitement comes from a few things. For one, power consumption characteristics which are expected to be almost Apple Silicon-like (tho single core PPW is likely to still be noticeably worse than M3). Then you have *most* benchmark results that have leaked which range from very encouraging to outright great, both in ST and MT considering the chips' TDP. It's also looking quite capable on the GPU side (tho we'll need to wait to see it in real games).

Those CB23 scores Jaykhin is getting are not great. They're in fact quite disappointing. But they're one data point compared to many and while I do think Cinebench is typically a decent benchmark it's also not the be-all and end-all of CPU benches. If LNL does end up being a poor CB platform, then that's one weakness it will have but it won't erase its many (expected) strengths.

 

[Nothingness]

The main excitement comes from the much increased efficiency/battery life in many real world usage scenarios like video playback or browsing or any other light workloads which is unheard-of for x86. Sure with 8 threads it won't set new efficiency records under full load but who does this constantly on a (low power) mobile device. If someone needs to do rendering tasks regularly he should better buy something with more cores and threads.

I see mac devices being used by many SW developers. And we need fast and efficient parallel builds. My 4 yo MBP can last all day running parallel builds and running mostly ST compute heavy tasks. It remains to be seen if Intel will be able to match that.

Is that a niche? Surely 😀

 

[majord]

Zen 5c is slower than regular Zen 5 by few %. The latter is 6% faster in Int. Regular Zen 5 is 10% faster than Zen 4 and if Skymont is 2% faster than RPL then the gap is essentially zero.

Since Skymont in Arrowlake clocks higher than Gracemont despite the P core sister losing frequency, I would wager on Skymont having significant clock advantage over 5C, if power is damned. It is possible if the 1.5mm+ size is true then maybe 30% extra core size is taken up to increase clock potential(which I think is stupid).

We have another example of very different approaches reaching a similar end result.

You're drawing a huge bow here I'm sorry.

2% faster than RPL in MTL to start with, and this is Intel's word. Hence why I said we need to wait for more data.
Zen 5 is more than 10% faster than Zen 4,
Zen 5c is the same architecture as Zen 5. any perceived difference in Strix application is L3 or latency related
you've ignored SMT
I already assumed a max clock difference between 5c and SKT , approx 4Ghz vs approx 4.6ghz