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.

 

[ToTTenTranz]

Battlemage is already competitive with that so if Celestial is a significant upgrade (or gets a new XeSS to compete with FSR4) then those 12Xe3 are going to make a meal out of anything RDNA 3.5.

Truth be told, Celestial doesn't even need to be a significant upgrade over Battlemage for Panther Lake H to beat 16CUs RDNA3.5.
Just from being +50% larger than 140V and having higher effective bandwidth from access to L3, the iGPU in PTL-H should already be significantly faster.

The iGPU on Medusa Point won't matter for the majority of people buying it.

The iGPU in Medusa Point was designed to not matter to anyone. It's a SoC for either office use or to be paired with a dGPU.

 

[511]

Truth be told, Celestial doesn't even need to be a significant upgrade over Battlemage for Panther Lake H to beat 16CUs RDNA3.5.
Just from being +50% larger than 140V and having higher effective bandwidth from access to L3, the iGPU in PTL-H should already be significantly faster.

But Celestial is a significant upgrade over Battlemage

 

[branch_suggestion]

And if Medusa Halo and Halo mini are positioned like Stryx Halo (so practically nothing uses them) and Medusa Point is the one that has to compete with PTL/NVL-H, then that's great for Intel iGPU wise considering Medusa Point will apparently stick with RDNA 3.5.

Medusa Halo will destroy Strix Halo by such a large margin that it should convince enough OEMs to cater for the new swimlane.
Trust me, it is not a pricing issue but entirely a commitment issue relating to having to invest in 256-bit platforms with what is technically the 3rd largest IHV in Windows PCs.
Many are holding back for N1X and then will decide if the extended memory thingie has a future.

Medusa Premium is a drop in replacement for Strix Point for premium T&L. It is far superior in every way.
If you want CPU perf instead you get Medusa Point with the CCD addon for green sticker laptops. Should have Strix Halo level MT perf.
If you want value you get Medusa Point by itself, which is the KRK replacement.

PTL has TTM on it's side and all the SKUs have room to shine, plus 18A means better cost structure.
NVL does not have the same luck, they cannot compete on value or volume either due to using N2 alongside AMD, so will purely be merit.
On the high end, NVL-HX cannot compete with 2 CCD Gator Range, that is a problem but 1 CCD Gator and Medusa Point w/CCD are there to battle.
NVL-AX if it does not get canned will still be DOA, I do kinda hope it does come out anyway as a proof of concept.
As for all the other configs, it all fits in the same cluster as the Medusa family and will come down to factors beyond pure performance.

Truth be told, Celestial doesn't even need to be a significant upgrade over Battlemage for Panther Lake H to beat 16CUs RDNA3.5.
Just from being +50% larger than 140V and having higher effective bandwidth from access to L3, the iGPU in PTL-H should already be significantly faster.

Oh how I wish Strix Point had MALL. Beating an intentionally hamstrung part with what is really 2023 IP is not really a flex.

The iGPU in Medusa Point was designed to not matter to anyone. It's a SoC for either office use or to be paired with a dGPU.

It brings near Strix Point iGPU perf into KRK pricing which is alright.

 

[511]

PTL has TTM on it's side and all the SKUs have room to shine, plus 18A means better cost structure.
NVL does not have the same luck, they cannot compete on value or volume either due to using N2 alongside AMD, so will purely be merit.
On the high end, NVL-HX cannot compete with 2 CCD Gator Range, that is a problem but 1 CCD Gator and Medusa Point w/CCD are there to battle.
NVL-AX if it does not get canned will still be DOA, I do kinda hope it does come out anyway as a proof of concept.
As for all the other configs, it all fits in the same cluster as the Medusa family and will come down to factors beyond pure performance.

That's not going to happen NVL-HX is no slouch in GPU Department.
For AMD vs Intel Volume we all know how that has happened over the years AMD won't have volume and by the time Medusa Halo has launched 18A would be producing more wafers than AMDs 2nm/3nm Capacity combined.
For CPU side 8+16/12C ccd ought to be enough for APU.

 

[511]

Oh how I wish Strix Point had MALL. Beating an intentionally hamstrung part with what is really 2023 IP is not really a flex.

Arc 140T is also a 2023 IP it's based on Xe1 and matches 890M

 

[branch_suggestion]

That's not going to happen NVL-HX is no slouch in GPU Department.

iGPU perf is irrelevant for -HX.

For AMD vs Intel Volume we all know how that has happened over the years AMD won't have volume and by the time Medusa Halo has launched 18A would be producing more wafers than AMDs 2nm/3nm Capacity combined.

NVL relies on N2 above all else, PTL sure.
if Intel thinks they can sell through all that 18A capacity, good luck to them.
AMD has got a lot of N2/N3 wafers lined up, they are expanding volume in line with their expanding market network, something that has been talked about recently that they are investing heavily into.

For CPU side 8+16/12C ccd ought to be enough for APU.

Sure, but with both MDS-H and Gator Range you have 24 full fat Z6 cores on mobile.
Intel not having a 2 tile on mobile means the closest comp is X Elite 2.

Arc 140T is also a 2023 IP it's based on Xe1 and matches 890M

Yeah because it is more or less a membw test, both are memory bound.

 

[511]

NVL relies on N2 above all else, PTL sure.
if Intel thinks they can sell through all that 18A capacity, good luck to them.
AMD has got a lot of N2/N3 wafers lined up, they are expanding volume in line with their expanding market network, something that has been talked about recently that they are investing heavily into.

it's only one N2 tile vs like AMD's whole design lining on TSMC with increased price it will still be less than Intel's wafer Fab 52/62 Combined capacity is 700K WSPY according to what Danniel Nenni comparing Capacity on Intel vs AMD is meaningless Intel will always win.

Qualcomm CEO: Intel is not an option today. We would like Intel to be an option

Qualcomm Inc. Chief Executive Officer Cristiano Amon said Intel Corp.’s production technology isn’t currently good enough for the maker of mobile phone processors to use as a supplier. If Intel is able to advance its manufacturing techniques to produce more efficient chips, then Qualcomm would...

semiwiki.com

Sure, but with both MDS-H and Gator Range you have 24 full fat Z6 cores on mobile.
Intel not having a 2 tile on mobile means the closest comp is X Elite 2.

You would be surprised but the 52C is coming to mobile

Yeah because it is more or less a membw test, both are memory bound.

all iGPUs are nowdays they don't get enough none gets enough maybe the KRK/4Xe3 Core are the only exception

iGPU perf is irrelevant for -HX.

Oh F i mistook it for AX lol my bad

 

[511]

The names have been getting out of hand from core i5/i7/i9 to to Ultra 5/7/9 and R5/R7/R9 to AMD AI HX 3700 wtf again and now with Halo
Ryzen AI MAX 395+

 

[ToTTenTranz]

Medusa Halo will destroy Strix Halo by such a large margin that it should convince enough OEMs to cater for the new swimlane.
Trust me, it is not a pricing issue but entirely a commitment issue relating to having to invest in 256-bit platforms with what is technically the 3rd largest IHV in Windows PCs.

The commitment issue is due to pricing. So it's a pricing issue.
OEMs have no problem committing to making PCBs with a 128bit SoC and a 128bit dGPU.

The problem was AMD trying to charge more for STX Halo than a high-end SoC and a mid/high-end dGPU (4070/5070 Mobile) combined.
If OEMs were going to charge the consumers >$2000 for a laptop that promotes its graphics and AI performance, they knew it wouldn't sell if it didn't have the green sticker. 99.9% of the people have no idea of AMD's RAM capacity advantage to run LLMs, and most of them use LLMs in cloud services anyways.

That AMD didn't understand this (nor consult with OEMs) before even planning to make and price this solution is the real failure here. And what a massive one at that.

Many are holding back for N1X and then will decide if the extended memory thingie has a future.

N1X is coming a year after STX Halo was made available. I don't think OEMs hold back a full year for a product range, it doesn't make sense for a company that releases products with a yearly cadence.

Medusa Premium is a drop in replacement for Strix Point for premium T&L. It is far superior in every way.

Sounds like AMD will try to price Medusa Premium well above Strix Point.
And unless there's some surprise in L2 cache amounts, LPDDR5X speeds (e.g. 12.7GT/s) or they simply wait for LPDDR6 availability, the iGPU performance advantage over Strix Point may be tiny.

 

[branch_suggestion]

The commitment issue is due to pricing. So it's a pricing issue.
OEMs have no problem committing to making PCBs with a 128bit SoC and a 128bit dGPU.

OEMs have been investing in that layout for decades.
256b soldered LPDDR has simply not been a thing.

The problem was AMD trying to charge more for STX Halo than a high-end SoC and a mid/high-end dGPU (4070/5070 Mobile) combined.
If OEMs were going to charge the consumers >$2000 for a laptop that promotes its graphics and AI performance, they knew it wouldn't sell if it didn't have the green sticker. 99.9% of the people have no idea of AMD's RAM capacity advantage to run LLMs, and most of them use LLMs in cloud services anyways.

That AMD didn't understand this (nor consult with OEMs) before even planning to make and price this solution is the real failure here. And what a massive one at that.

That is pure speculation.

N1X is coming a year after STX Halo was made available. I don't think OEMs hold back a full year for a product range, it doesn't make sense for a company that releases products with a yearly cadence.

Nah they are tentative because they
1. Don't trust AMD to ship sufficient volume or provide sufficient support to the OEMs through engineering or MDF which ends up as a negative feedback loop
2. NV are more than capable of coercing OEMs into waiting for their 256b part instead to start their market push nullifying AMDs first mover advantage

Sounds like AMD will try to price Medusa Premium well above Strix Point.
And unless there's some surprise in L2 cache amounts, LPDDR5X speeds (e.g. 12.7GT/s) or they simply wait for LPDDR6 availability, the iGPU performance advantage over Strix Point may be tiny.

Well unless they cut the cache down per UMC for AT3/4 it should be 16MB L2 for AT4.
It should be closer to Strix Halo iGPU perf than Strix Point.
As for price it will be a step above Strix Point, that is without question or it is a margin diluter.
But it really is a class above in all aspects, nobody else really has a direct competitor, so I expect probably Max 385 kinda pricing.

 

[511]

OEMs have been investing in that layout for decades.
256b soldered LPDDR has simply not been a thing.

That is pure speculation.

Nah they are tentative because they
1. Don't trust AMD to ship sufficient volume or provide sufficient support to the OEMs through engineering or MDF which ends up as a negative feedback loop
2. NV are more than capable of coercing OEMs into waiting for their 256b part instead to start their market push nullifying AMDs first mover advantage

Well unless they cut the cache down per UMC for AT3/4 it should be 16MB L2 for AT4.
It should be closer to Strix Halo iGPU perf than Strix Point.
As for price it will be a step above Strix Point, that is without question or it is a margin diluter.
But it really is a class above in all aspects, nobody else really has a direct competitor, so I expect probably Max 385 kinda pricing.

Not happening on a 128bit bus it's impossible if you don't have enough mem bandwidth for AT3

 

[ToTTenTranz]

OEMs have been investing in that layout for decades.
256b soldered LPDDR has simply not been a thing.

256bit LP5X is less of a problem than 128bit LP5X + 128bit GDDR6.
There's no reason to make such a big fuss about 256bit LPDDR5X. Apple's been doing it in volume production for 5 years.

That is pure speculation.

Price, or yet AMD's excessive bullishness on perceived STX Halo value is the only thing keeping OEMs. Just look at @adroc_thurston prideful comments on how great it was that STX Halo was being sold for so much money, since late last year.
Hubris ended up being the chip's demise.

It should be closer to Strix Halo iGPU perf than Strix Point.

AT4 with STX Halo performance? No way. Not in rasterization, at least and not at ~25-30W. Perhaps at 15W.

As for price it will be a step above Strix Point, that is without question or it is a margin diluter.
But it really is a class above in all aspects, nobody else really has a direct competitor, so I expect probably Max 385 kinda pricing.

DOA pricing.

 

[511]

Hang on its getting confusing how many FP32 ALU are we talking about same as PTL 12Xe3 Cores I presume so 1536 FP32 ALU for AT4 since it's a 24CU.

NVL Xe3P is also 12Xe3P cores.

 

[branch_suggestion]

256bit LP5X is less of a problem than 128bit LP5X + 128bit GDDR6.
There's no reason to make such a big fuss about 256bit LPDDR5X. Apple's been doing it in volume production for 5 years.

One it is MOP, two it is Apple, it's own vertically integrated ecosystem.
Not comparable to tight margin Windows OEMs.

AT4 with STX Halo performance? No way. Not in rasterization, at least and not at ~25-30W. Perhaps at 15W

STX-H will pull away at higher TDP, which is the point.
MDS-P is a 45W kinda part, perf does not need to scale higher than whatever it hits at cTDP.
It is 12WGP equiv/1SE RDNA5 vs 20WGP/2SE RDNA3.5, MDS-P will have over half the membw and that 16MB L2 should probably outperform 32MB of MALL.

DOA pricing.

Nah, 32GB Max 385 is not that much more expensive than 370HX from the limited sample size there is.
MDS-P should be outright faster outside of stuff that only cares about membw.

Hang on its getting confusing how many FP32 ALU are we talking about same as PTL 12Xe3 Cores I presume so 1536 FP32 ALU for AT4 since it's a 24CU.

NVL Xe3P is also 12Xe3P cores.

Correct (3072 w/VOPD).

 

[511]

Correct (3072 w/VOPD).

WHAT is VODP

 

[igor_kavinski]

WHAT is VODP
View attachment 129801

Search VOPD here: https://chipsandcheese.com/p/microbenchmarking-amds-rdna-3-graphics-architecture

 

[adroc_thurston]

Hubris ended up being the chip's demise.

Idk chefe it's doing aight and had a successor.
Mind need your eyes checked.

 

[regen1]

don' matter, it's gonna get roadkilled by the gfx13 blob either way.

We will see then how well they each do, little bit long to go.
Also with Intel volume is another thing. Hardly anything(in recent years) Intel volume at iGPU level can be termed 'roadkilled'.

 

[adroc_thurston]

Also with Intel volume is another thing

12 core iGPU parts are a meme for claiming leadership.

Intel volume at iGPU level can be termed 'roadkilled'.

Yeah man sure GT3/4e sold gangbusters right?

 

[ToTTenTranz]

Idk chefe it's doing aight and had a successor.

Doesn't look like it's doing alright with the negligible amount of laptop design wins it got. Even Kaby Lake G got more laptop design wins than STX Halo so far, which is meme worthy at this point.
As for the successor, well there's no successor yet and a couple months ago word was that its successor had been canceled.

If anything, it looks like RDNA4's success and AMD's faith in RDNA5 are what saved Medusa Halo, and not STX Halo's lack of success.

Nah, 32GB Max 385 is not that much more expensive than 370HX from the limited sample size there is.

Sources?

 

[jpiniero]

If anything, it looks like RDNA4's success and AMD's faith in RDNA5 are what saved Medusa Halo, and not STX Halo's lack of success.

More like Ayyy Eyeeee

 

[regen1]

12 core iGPU parts are a meme for claiming leadership.

Intel's ultra 5 are not that heavily cut-down for IGPUs as AMD's have been till now.
Ultra series 1&2 ultra 7/9 H : 8Xe1/+ , Ultra 5 H: 7Xe1/+
Panther will have Ultra 5 SKUs with 10Xe3.

We will see then how well they each do, little bit long to go.
Also with Intel volume is another thing. Hardly anything(in recent years) Intel volume at iGPU level can be termed 'roadkilled'.

Yeah man sure GT3/4e sold gangbusters right?

GT3/4e is recent to you ?
Either way Intel's IGPU volume in notebooks is completely another level to AMDs.

 

[adroc_thurston]

Panther will have Ultra 5 SKUs with 10Xe3.

Hahahahahahaha hahahaha oh yes man

 

[adroc_thurston]

Doesn't look like it's doing alright with the negligible amount of laptop design wins it got.

It's a new swimlane indeed.

Even Kaby Lake G got more laptop design wins

Way back when Intel had 99% of premium share.

If anything, it looks like RDNA4's success and AMD's faith in RDNA5 are what saved Medusa Halo, and not STX Halo's lack of success

It's the opposite, low-end discrete dGFX is now made of reused iGPUs.
AMD surrendered in discrete graphics.

 

[Doug S]

N1X is coming a year after STX Halo was made available. I don't think OEMs hold back a full year for a product range, it doesn't make sense for a company that releases products with a yearly cadence.

You underestimate the power of the Nvidia branding when it comes to AI. That is something OEMs would consider well worth waiting for.