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.

 

[DavidC1]

Skymont acheived crazy SPEC uplifts because the previous e-core was meh

You have to separate into Int and FP for more accurate comparisons. Primate Labs have regressed since GB4. With GB4 it was separated into both, and had proper sorting functions. They've been following the mobile trend of simplifying everything to the point of being dumb.

There's a reason that Scalar Integer is the prize for CPU uarchs. That's the real hard part to improve. It takes real innovation to make a top uarch. FP/Vectors are much easier to improve. Also speeding up Scalar Integer boosts everything, emulation, games, FP, AI. It's a foundation. It's the same reason why Apple can make a CPU that's absolutely class leading but doesn't do the same in GPU. GPUs need smarts too, but you can brute force lots of them and it's heavily in mercy to Moore's Law advancements. You cannot do that with CPU.

Also, I find it funny people call 30% difference meh, when we frequently argue over 5% differences. I knew since Goldmont Plus that the E core team had potential. They are fundamentally different from the P core team. I hope they continue rather than being "infected" with P core philosophy.

 

[DavidC1]

There is a reason to stop at 512b. Picking bigger would have far reaching consequences for the rest of the chip. It's quite obvious if you think in Bytes rather than bits.

It should have stopped at 256-bits. From now the Moore's Law gains are crashing. Going from 100% density gains to 30%. At one point it'll be like everything non-computers, where advancements come hard, expensive, and slow. Like in architecture, energy storage/generation.

 

[poke01]

The E core is to eventually replace the P core that is why I'm harsh. I can see the limitations of the P core. What the Austin team did with the previous monts is amazing and I'm not dismissing but having the absolute perfromance crown speaks volumes and skymont is one step closer to that reality.

Also, I find it funny people call 30% difference meh, when we frequently argue over 5% differences. I knew since Goldmont Plus that the E core team had potential. They are fundamentally different from the P core team. I hope they continue rather than being "infected" with P core philosophy.

 

[DavidC1]

The E core is to eventually replace the P core that is why I'm harsh. I can see the limitations of the P core. What the Austin team did with the previous monts is amazing and I'm not dismissing but having the absolute perfromance crown speaks volumes and skymont is one step closer to that reality.

It absolutely will, because the P core requires tremendously more xtors to achieve not much. Gaming comparisons being so close is embarassing, because it's hard to multi-thread and is latency sensitive, but the two uarchs are within 5% and Lion Cove needed higher clocks to do so. The tiny difference is what allows the 1P+16E config to beat the base config or 8P config in some gaming scenarios. It's close enough.

Also, when I say Skymont is amazing, it's still relative to x86. They could do FAR better. Phone chips being anywhere near in performance is a total embarassment to x86. We shouldn't even question this, because it's that bad for x86. Go back to AT 15 years ago and say future iPhone chips will beat 99% of same year desktop chips. When people say x86 as an ISA sucks, they are mostly talking about decode. People say the clustered decode in Skymont *solves* that problem. Actually, I agree. it's better perf-wise than a monolithic decoder. It's absolutely no "cheap wide decode". Intel manual says that it will allow adding more decode width linearly.

The team has been bound by 1mm2 core area. Let's say they aim for 2mm. What would that do?

Now let's see them do *real* things. Threaten ARM.

 

[gdansk]

the anti-Intel waves that some are caught in

I do not have to admire it nor follow the delusions of Skymont messianic cult. It's Zen 4 but without the ability to clock high nor support AVX-512 instructions a few years later on a better process. Amazing. Wow. And they perpetually insult Lion Cove too. But on desktop parts Skymont is a waste of silicon at present. It's the most over-hyped mediocrity. Because in a few years a core that has some resemblance to it will save Intel?

Nope, I don't care. For desktop they have to do better. Hopefully another die shrink and throwing more silicon at the problem can make it something interesting.

 

[Covfefe]

If you read the article with attention, you would see that Skymont is shown as somewhat competitive with Zen4, then Zen2 comparison is made to highlight what it is weak at. Skymont is much bigger core (I don't mean mm^2), it has bigger ROB, INT/FP Reg files, more execution units, wider decoder and wider rename. It also probably can sustain higher average clock and cluster vs cluster has more L2. Comparing it to Zen2 for most of the article would be a waste of time.

If your read their Zen5 focused articles they often mention too little of L1i, slow frontend (I mean 4 cycles for a function call, c'mon), too small int register file. Should we say it's odd too?

Thanks, I wasn't reading with attention before.

I disagree that only showing specific benchmarks where Zen2 wins is the best way to cover Skymont's strengths and weaknesses. Including the 3950X in the main SPEC summary charts would've offered far more context. They could still discuss how poorly Skymont performs in SIMD benchmarks, but they have the added benefit of showing how much Skymont, Zen4, and Zen5 beat Zen2 in the rest of the SPEC suite.

Since you brought up their treatment of AMD, here is how they talk about Zen5 in the conclusion to their "AMD’s Ryzen 9950X: Zen 5 on Desktop" article. Positive points highlighted in green, negative in red.

Spoiler

AMD deserves credit for how fast they’ve been able to iterate. Since Golden Cove launched in 2021, AMD has released Zen 4 and Zen 5. Both bring significant architecture changes. On Intel’s side, Raptor Cove reuses Golden Cove cores at higher clock speeds with more cache. Redwood Cove does have smattering of minor improvements like more aggressive prefetching, a larger micro-op queue, and doubled L1i capacity. But major structure didn’t see layout or capacity changes, so Intel’s changes have minor impact compared to AMD’s. And Zen 5’s improvements can be quite impressive in areas once dominated by Intel, like AVX-512.


As for AVX-512, AMD’s engineers didn’t stop at creating one strong FP/vector unit. Instead, they designed, optimized, and validated two versions of it to better fit mobile and desktop requirements. That was done alongside implementing the core in performance and density optimized variants. Add on significant core changes, and it’s clear Zen 5 took a lot of engineering bandwidth. For sure, AMD still economizes engineering effort by reusing Zen 4’s IO die and not maintaining two completely different core architecture lines as Intel does. But AMD today feels more ambitious and capable than it was in 2017.


But any engineering team has limits, and compromises are inevitable considering how fast AMD is pushing out cores. Zen 5’s integer register file didn’t get large enough. Clock speed increases are minor compared to prior generations. AMD’s first clustered decode implementation can’t have both clusters work on a single thread. Widening the core may have been premature too. Much of the potential throughput offered by Zen 5’s wider pipeline is lost to latency, either with backend memory accesses or frontend delays.

Yep, file compression is heavily latency limited too, with some spicy challenges for the branch predictor mixed in
Fortunately, some of those compromises have little impact. 4-wide decode likely isn’t a limiting factor for single threaded performance, thanks to high micro-op cache hitrates and other limiting factors coming first. Minimal clock speed gains are paired with lower power consumption. And when paired with faster memory, the Ryzen 9 9950X can comfortably beat the Ryzen 9 7950X3D in productivity workloads.
As a foundation, Zen 5 looks solid and I’m excited to see how AMD builds on it in the future. I’m also looking forward to VCache-enabled Zen 5 variants. Increasing L3 capacity from 32 to 96 MB on Zen 3 and Zen 4 did wonders for certain games. Its largest issue was losing clock speeds, which made it a hit or miss affair on other programs. If Zen 5 can keep clocks up, a VCache variant that loses less core width to memory latency could be a force to be reckoned with.
Again, we would like to thank AMD for sending us over some Zen 5 CPUs for review and if you like our articles and journalism, and you want to support us in our endeavors, then consider heading over to our Patreon or our PayPal if you want to toss a few bucks our way. If you would like to talk with the Chips and Cheese staff and the people behind the scenes, then consider joining our Discord.

 

[DavidC1]

I do not have to admire it nor follow the delusions of Skymont messianic cult. It's Zen 4 but without the ability to clock high nor support AVX-512 instructions a few years later on a better process. Amazing. Wow. And they perpetually insult Lion Cove too. But on desktop parts Skymont is a waste of silicon at present. It's the most over-hyped mediocrity. Because in a few years a core that has some resemblance to it will save Intel?

Nothing to do with "messianic" for me. Just extrapolation based on history. Same as Pentium M vs Pentium 4. Enough history suggests the E core team is far more capable at substantial core changes without falling flat on their face.

 

[gdansk]

Nothing to do with "messianic" for me. Just extrapolation based on history. Same as Pentium M vs Pentium 4. Enough history suggests the E core team is far more capable at substantial core changes without falling flat on their face.

It will prove delusional. The ILP wall comes for all its descendants all the same. And at the end of the day Skymont will always be Skymont, not its children.

 

[Covfefe]

Nothing to do with "messianic" for me. Just extrapolation based on history. Same as Pentium M vs Pentium 4. Enough history suggests the E core team is far more capable at substantial core changes without falling flat on their face.

Yep, same as Zen1 was back in 2017, or Qualcomm X Elite is now. The design team has shown that they're able to execute. The core architecture is solid. And the CPU core already has clear, quantifiable benefits over it's competitors.

 

[DavidC1]

It will prove delusional. The ILP wall comes for all its descendants all the same.

Yea, and back in Athlon 64 days some expressed doubts whether they could make more capable core.

Skymont is far, far away from the ILP wall, and that's set by ARM camp including Apple. I have doubts whether Apple is suffering from some sort of complacency especially after GWIII departure too. The biggest clue is that even from ARM perspective Skymont is small. All the "wide" parts of the core resulted in that small of a size, meaning compromises are made. For example, only 2 of the 8 ALUs are capable of executing all instructions. Rest are simple or to quote the lead "they were in cheap in area to add". In branch prediction, it's on par with Lion Cove but still behind Zen 5 in accuracy and BTB sizes.

Or, you also believe in the ARM "magic".

And at the end of the day Skymont will always be Skymont, not its children.

No true high performance core is built entirely from the ground up. It still started from Bonnell Atom, which itself started from expanding P54C, the original Pentium.

 

[gdansk]

Yea, and back in Athlon 64 days some expressed doubts whether they could make more capable core.

And lo, the naysayers were misconstrued yet again. The diminishing returns hits all cores. Some more than others, but a mont-like Unified Core will still be hoisted by its own universalist petard. In catering for the server market it will be less optimal.

Cove might not be great but it has its advantages. There are technical (i.e. non-political or managerial conflict) reasons it was kept alive. Will you even admit that? I don't know. And I'm not going to touch on comparisons to ARM since that's an apple in this orange thread. When the world recompiles its software frequently it obtains a "better" wall for such efforts.

 

[DavidC1]

And lo, the naysayers were misconstrued yet again. The diminishing returns hits all cores. Some more than others, but a mont-like Unified Core will still be hoisted by its own universalist petard. In catering for the server market it will be less optimal.

That's why you need a rethink too once in a while, and doesn't always come from your team, or even your company. This is also why collective mindset can be a big penalty, because everyone gets pigeonholed into one thinking, until someone else comes and upends it entirely.

Cove might not be great but it has its advantages. There's a technical non-political reasons it is kept alive. Will you even admit that? I don't know. And I'm not going to touch on about comparisons to ARM since that's an orange in this thread. When the world recompiles its software frequently it obtains a "better" wall for such efforts.

It's kept alive because it's easier to keep and that's what they've been developing on it all the time. There are many, many cases where you don't move to a superior place because it's not worth the time/effort for you. But there is a tipping point.

In catering for the server market it will be less optimal.

Yea, a power efficient/area efficient core is one area where server fits well.

 

[gdansk]

There are many, many cases where you don't move to a superior place because it's not worth the time/effort for you.

Moving to an imagined superior point is a great idea but in this case counterfactual. LCove improved performance* in more workloads over RC. Skymont didn't so it doesn't get to be the big core yet. It isn't complicated. The rest is simply projections of a messianic tech cult.

We need Intel to be better and I don't give a crap how they get there. If it looks more like mont or Cove I do not care. But I suspect a lot more of Cove will survive in UC than you suspect and that's fine. Pragmatically there is plenty to pick from both lineages.

 

[511]

https://twitter.com/x/status/1959543923694526506

 

[igor_kavinski]

Hmm...JEDEC must have an upcoming DDR5-7200 spec.

 

[511]

PTH171/PTH173ï½Intel®ï½Industrial Motherboardsï½DFI

Mini-ITX, Intel Next-generation of Core Ultra processor, DDR5, 1 DP++, 1 HDMI 2.0, 1 USB Type C, 1 M2A-Display extension port, 1 PCIe x4, 1 M.2 M Key, 1 M.2 B Key,1 M.2 E Key, 1 M.2 A Key, SIM, 3 Intel 2.5GbE, 2 COM, 4 USB 3.2 Gen2 Type A, 1 USB Type C, 5 USB 2.0, 2 SATA 3.0, Audio, EXT-OOB, DC-Jack

www.dfi.com

 

[igor_kavinski]

PTH171/PTH173ï½Intel®ï½Industrial Motherboardsï½DFI

Mini-ITX, Intel Next-generation of Core Ultra processor, DDR5, 1 DP++, 1 HDMI 2.0, 1 USB Type C, 1 M2A-Display extension port, 1 PCIe x4, 1 M.2 M Key, 1 M.2 B Key,1 M.2 E Key, 1 M.2 A Key, SIM, 3 Intel 2.5GbE, 2 COM, 4 USB 3.2 Gen2 Type A, 1 USB Type C, 5 USB 2.0, 2 SATA 3.0, Audio, EXT-OOB, DC-Jack

www.dfi.com

Good work, detective!

 

[511]

Good work, detective!

i found it on twitter lol

 

[511]

View attachment 129149View attachment 129150

https://twitter.com/x/status/1959543923694526506

upgradable DDR5 7200 is rare especially JEDEC

 

[igor_kavinski]

So they are gonna sell it to the military for drones!

Kind of a waste since they get destroyed so much.

 

[511]

So they are gonna sell it to the military for drones!

Kind of a waste since they get destroyed so much.

You missed the important part according to this PTL is H2 25 Launch

 

[DavidC1]

Moving to an imagined superior point is a great idea but in this case counterfactual. LCove improved performance* in more workloads over RC. Skymont didn't so it doesn't get to be the big core yet. It isn't complicated. The rest is simply projections of a messianic tech cult.

That's similar to sticking to Netburst for a while. It was still the highest performance despite the negatives. Also up until Skymont vast majority still were hoping on Cove. I bet a significant still are.

And I meant going to a bigger mont soon rather than later and replacing the main one with Stephen Robinson's team.

 

[gdansk]

That's similar to sticking to Bulldozer or Netburst for a while.

Nope, it's sticking with the higher performance core.
It's actually, really, truly the opposite of what you claim and that's the main reason I am so very confused by the Skymont is messiah cult. It isn't. Maybe something in that line may be but it is not. No amount of counterfactual imagination will make Skymont higher performance than LC today.

 

[DavidC1]

Nope, it's sticking with the higher performance core.
It's actually, really, truly the opposite of what you claim and that's the main reason I am so very confused by the Skymont is messiah cult. It isn't. Maybe something in that line may be but it is not Skymont.

No one said it's Skymont, you are the first one.

I said the E core base and team is. So do most people.

 

[gdansk]

No one said it's Skymont, you are the first one.

I said the E core base and team is. So do most people.

And future parts from that team will not make Skymont better.
I'm glad we can agree that Skymont is really not worth talking about. Well, for desktop use. It is good enough elsewhere.