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.

 

[Exist50]

WTFTech will post anything they get their hands on. Some of it might be legit.

The Youtubers in general just make up stuff.

Nah, they're equally as terrible. Both pull the "make stuff up then delete what doesn't come true" schtick. Good examples:

A Troubling Trend

WHAT?! Guru3D can no longer consider them selves a legit new source? Oh how the mighty have fallen.

www.overclock.net

https://www.reddit.com/r/Amd/comments/6bih5e

 

[mikk]

Thanks. Is Arrow Lake going to use the same P core uarchitecture as Meteor Lake but on a different node?

No they differ. Redwood Cove and Lion Cove.

 

[A///]

if the slides are real then email feels confident about the new p and e core architectures, and it is also possible intel has figured a way to enable ht on those e cores. An 8+16 lay out will be 24 cores 48 threads. Intel's chiplet approach reduces their bom and manufacturing errors, learning something from amd. would it be possible for intel to design e cores that are as powerful as a zen 4 core or raptorlake p core? yes.

intel is the beluga whale of semi cons. it's spent the last 5 years being beaten by amd and rolling over for belly rubs, now it's the killer whale hunting an innocent baby seal. chomp chomp

 

[igor_kavinski]

would it be possible for intel to design e cores that are as powerful as a zen 4 core or raptorlake p core? yes.

now it's the killer whale hunting an innocent baby seal. chomp chomp

Someone is suffering from DELUSIONSSSSSS

 

[msj10]

Someone is suffering from DELUSIONSSSSSS

Not really. the difference in IPC between gracemont and zen 4/golden cove is around 30-40%. it's not unbelievable that they can reach that after a couple of generations.

 

[A///]

Someone is suffering from DELUSIONSSSSSS

the same was said about amd making waves after zen 1. not before but after. intel has finally learned they cannot throw money at a problem to solve it and need to work cohesively together to get the problem solved as amd has.

 

[A///]

for the record I'm neither pro intel or amd. but it's obvious amd were blindsided by raptor lake when the leaks weeks and months before made it look like nothing. @Markfw and i had many a chat wondering why intel was going forward with a refresh that was looking to be at tops 3-5% better over alderlake. I was very smug on the morning of raptorlake's release expecting intel fans to lose their collective minds. instead my jaw needed to be picked up off the floor. I haven't seen a finely tuned raptorlake system because no one fine tunes their systems anymore save for other old farts and i haven't got the patience for that stuff anymore. manually tuning your ram vs xmp is better, but most people can't or don't want to learn about ram tuning.

amd has a big challenge ahead of themselves in the next 16 months to deliver zen 5 and make it better than whatever lake crap intel is pushing out then.

 

[Markfw]

for the record I'm neither pro intel or amd. but it's obvious amd were blindsided by raptor lake when the leaks weeks and months before made it look like nothing. @Markfw and i had many a chat wondering why intel was going forward with a refresh that was looking to be at tops 3-5% better over alderlake. I was very smug on the morning of raptorlake's release expecting intel fans to lose their collective minds. instead my jaw needed to be picked up off the floor. I haven't seen a finely tuned raptorlake system because no one fine tunes their systems anymore save for other old farts and i haven't got the patience for that stuff anymore. manually tuning your ram vs xmp is better, but most people can't or don't want to learn about ram tuning.

amd has a big challenge ahead of themselves in the next 16 months to deliver zen 5 and make it better than whatever lake crap intel is pushing out then.

Well, I am not surprised. Exactly. It may be close to Zen 4 in performance, but the power output at stock is insane ! And my 7950x's are set at -25 CO, and take 150 watt and beat most ANY Raptor benchmarks at anything below 250 watts.

 

[A///]

Well, I am not surprised. Exactly. It may be close to Zen 4 in performance, but the power output at stock is insane ! And my 7950x's are set at -25 CO, and take 150 watt and beat most ANY Raptor benchmarks at anything below 250 watts.

The power is insane, but only if it's used for production. that's the caveat innit? That if you are buying that chip, you should be doing production and not gaming. otherwise the 13700k is the better buy and only 2pc less in gaming performance. and it can be oc'd and perform as well as a 13900k. the cost savings are there for those who can do it. both the 13900k and 7950x need hours of fine tuning to make them perform better than stock. roman's video on taming the 13900k showed just that, but he didn't share all the values he used and how long it took him to get there. his p/e is deadlocked with a 7950x but he may have spent 2 weeks tuning his system. people keep linking to his video as a means of saying yeah it can be tuned and use less power, but keep in mind most of these people can't tell the difference from their bum to their head.

When I was running a quad duo I spent an agonizing month getting everything tuned up perfectly just to get 20-25 fps more out of games. was it worth it? you bet your patootie it was. are most of these kids going to do it? Nah.

 

[A///]

someone mentioned it a few years ago when rocket sink lake came out but alderlake seemed like a pipe cleaner job to get something out and get user data off of to then improve on raptor. It's not a pipe clean job like on a new node but the theory has some weight to it. The slides from earlier make me think mtl is a pipe clean to get the mcm setup for intel going while alderlake coming later on will be their meaty offering that's been fixed. Royal core is allgedly integrated into the cores coming on arrow and lunar or the other non mobile one.

 

[Henry swagger]

Not really. the difference in IPC between gracemont and zen 4/golden cove is around 30-40%. it's not unbelievable that they can reach that after a couple of generations.

They just need the e-cores to hit 5.5 or 5.8 ghz all cores on intel 4 from 4.6ghz on 7nm

 

[DrMrLordX]

They just need the e-cores to hit 5.5 or 5.8 ghz all cores on intel 4 from 4.6ghz on 7nm

That kind of defeats the purpose of e-cores.

 

[A///]

it'll come down to architecture for the upcoming p and e cores. if they can manage to get ht working on them e cores and that means working with microsoft to get their windows scheduler working correctly, then it can be a fem fatale of a processor series. The obsession with speed is eons old. we've seen time and time again in x86 history where speed does not equal better performance. I'm interested in how intel's leaked slides pan out if they're true or not and bait for amd and how amd will design zen 5 and address their lack of cores. call e core spam useless as much as you want but they're still cores at the end of the day. The lack of hyper threading on them is hurting them, but this has to do with many intel microsoft issues most of which is getting them to work correctly and power and heat envelopes.

Now that we're mostly out of the pandemic we can witness amd return to their previous covid cadence of 12-16 months. There was a 22 month difference this last time around. I remember our dear @DrMrLordX stating he suspected zen 4 would be finished long before their launch if it weren't for it. In a non covid world these processors would have come out months ago. Intel has maintained their cadence thanks to their near vertical setup. although im not sure how much sales amd would have accrued by this time. coupled with high costs, amd made a flawed call in their presentation months ago to mention vcache processors. now everyone is waiting for those. impatient and those with deep pockets bought in already and will upgrade to vcache the day it drops.

 

[IntelUser2000]

So reddit rumors were wrong and MLID is just to be ignored at this point.

You know who's a good leaker? Raichu.

Not really. the difference in IPC between gracemont and zen 4/golden cove is around 30-40%. it's not unbelievable that they can reach that after a couple of generations.

The difference between the two is ~50%. Yes it's plausible by Arrowlake timeframe, Skymont will perform like Golden Cove, or likely even better but Lion Cove will perform faster still.

I do believe the gap between them will decrease though, which will silence the "E core spam" critics because it'll become easier to manage the two plus scheduling will improve.

The stopping of E cores at 16 starts to sound like how they expected mobile CPUs to quickly scale to 32+ cores, and it pretty much stopped at 8. Like, these people who are in charge of graphs needs to be put down a notch at least. All they think is everything will scale the way it scaled before, and linearly. Similar minded people were put in charge of "GHz wars" at Intel.

I don't expect Hyperthreading to be enabled on the E cores. It just doesn't make as much sense. I think the big.little concept is limited, where you use the little cores for lowering power use in light load. I think this concept of maximizing performance has lot to be tapped into.

You further specialize the cores for maximum performance. Apple seems to be taking one approach, where since A10(big.little) the performance of the big cores have went up drastically, but the little cores are still meant for power saving. It should be a two-in-one punch of maximizing P cores for low thread count and maximizing E cores for high thread count.

The way they can use the hybrid concept to maximize performance is numerous. It's a low hanging fruit.

 

[IntelUser2000]

So in the mobile Meteorlake leak, it showed U chips having up to 12 cores. That's a 4+8 config, meaning it's twice the amount of P cores compared to Alderlake U's 2+8.

 

[poke01]

So in the mobile Meteorlake leak, it showed U chips having up to 12 cores. That's a 4+8 config, meaning it's twice the amount of P cores compared to Alderlake U's 2+8.

MTL will be great in mobile. Arrow Lake will be excellent in mobile. I am not excited for desktop anymore. It's more fun to see what Intel/AMD do in thermal constrained devices.

As efficiency becomes more important mobile will reap the biggest benefit this decade.

 

[Geddagod]

don't expect Hyperthreading to be enabled on the E cores. It just doesn't make as much sense. I think the big.little concept is limited, where you use the little cores for lowering power use in light load. I think this concept of maximizing performance has lot to be tapped into.

It depends on the amount of extra performance they can get from HT, I think. I'm no computer architect, but I believe different architectures have different amounts of performance they gain from adding HT, but I think the average is 20-30%. So if the die size/power cost is higher than that, I doubt they will implement it, since the entire point of E-cores are efficiency. But I think the most important factor that will play into this is die area cost, since e-cores seem to be most maximized for area efficiency than anything else.

 

[Doug S]

That kind of defeats the purpose of e-cores.

Not really, it depends on why they are there. Intel's E cores seem to be primarily about saving area, not saving power. If their goal was maximizing performance per watt they did a pretty bad job of it, but they did a heck of a job of maximizing performance per area, to allow more performance for a certain chip size for servers and the kind of high end PCs that use server like CPUs with a ton of cores that run flat out for long periods.

 

[Geddagod]

Not really, it depends on why they are there. Intel's E cores seem to be primarily about saving area, not saving power. If their goal was maximizing performance per watt they did a pretty bad job of it, but they did a heck of a job of maximizing performance per area, to allow more performance for a certain chip size for servers and the kind of high end PCs that use server like CPUs with a ton of cores that run flat out for long periods.

It's cuz they clocked the e-cores so high

 

[dullard]

but it's obvious amd were blindsided by raptor lake when the leaks weeks and months before made it look like nothing. @Markfw and i had many a chat wondering why intel was going forward with a refresh that was looking to be at tops 3-5% better over alderlake.

Why on earth would you think that something with 30% to 50% more cores (albeit E cores), a tweaked core design, almost 10% faster turbo, 17% faster memory, and more cache would be just 3% to 5% faster? Heck the 5% higher turbo power level alone would let Raptor Lake be more than 3% faster than Alder Lake even if you ignored all the other changes.

I've always been in the camp that Raptor Lake would be much better than Alder Lake but Meteor Lake is going to disappoint many people (wait for Arrow Lake). Unless someone selectively picks only bad rumors and not any of the good rumors, then there had to be a much more significant speed bump than "at tops 3% to 5%".

Always take pre-release benchmarks with a large grain of salt. Your 3% to 5% number can be changed with a BIOS update alone.

 

[dangerman1337]

The problem of no 8P SKU potentially with Meteor Lake desktop is that Intel has made a big deal with Thread Director with Raptor & Alder Lake, I mean I know Meteor Lake E cores are going to be way better but would they not have say an 8P+8E 14700K that bests a 7800X3D?

 

[dullard]

The difference between the two is ~50%.

IPC varies quite a lot depending on the workload. ChipsAndCheese did a nice analysis over a variety of workloads and came up with:
"IPC over the entire workload was 1.72, 1.91, and 2.25 for Gracemont, Zen 2, and Golden Cove respectively"

Gracemont: Revenge of the Atom Cores

This article can be considered a part 2 to our Golden Cove article because today we are looking at the other core in Alder Lake, Gracemont.

chipsandcheese.com

2.25 / 1.72 = 1.308 (30.8% higher IPC for Golden Cove vs Gracemont).

 

[mikk]

The problem of no 8P SKU potentially with Meteor Lake desktop is that Intel has made a big deal with Thread Director with Raptor & Alder Lake, I mean I know Meteor Lake E cores are going to be way better but would they not have say an 8P+8E 14700K that bests a 7800X3D?

MTL-S most likely won't come 2023, it wouldn't help against Zen 4 3D and I don't think this is a problem because Zen 5 won't come before 2024 either.

MTL-S limited to 6P cores makes sense when ARL-S launching at the same time or almost at the same time on the same platform. It's not the typical release cycle with 1 year difference. Both Raichu and kopite7kimi implied this.

https://twitter.com/x/status/1533799277763657729

https://twitter.com/x/status/1529670185938198528

 

[IntelUser2000]

Not really, it depends on why they are there. Intel's E cores seem to be primarily about saving area, not saving power.

I think we're still too early in the hybrid era to conclude this. Future architectures will likely better this so the E core will include efficiency, not just die area.

Look at how they significantly improved the ring frequency problem with Raptorlake, something that people saw it as an issue of having an E core.

2.25 / 1.72 = 1.308 (30.8% higher IPC for Golden Cove vs Gracemont).

I disagree, that's too low.

Gracemont in Alderlake = Slightly below Skylake
Gracemont in Raptorlake = Little better.

Sunny Cove = 18% over Skylake
Golden Cove = 19% over Sunny Cove

This assumes zero gain for Willow Cove. Situationally the larger cache makes a difference. 1.18 x 1.19 = 40.4%. Now if Raptorlake's GMT is still slightly below Skylake, 3% gets us to 44%. If you assume Golden Cove is only 30% faster than Gracemont, then you have to assume Golden Cove is less than 30% faster than Skylake. That's not the case, we agree?

The architecture(basically "quality" of integer performance) of Gracemont itself might be on par Skylake, but the caches and buses are not, and it's weaker in floating point. In real world applications, that'll make it worse.

Meteor Lake is going to disappoint many people

I mean I know Meteor Lake E cores are going to be way better but would they not have say an 8P+8E 14700K that bests a 7800X3D?

Hardly an issue if Arrowlake, a much better product comes in "almost at the same time" and be labelled the same generation. I can expect everything above xx600K to be Arrowlake for example. Actually considering 13700K is 8+8, it can replace even that. But since gamers are very finicky bunch, better they have Arrowlake just for K chips.

 

[dangerman1337]

MTL-S most likely won't come 2023, it wouldn't help against Zen 4 3D and I don't think this is a problem because Zen 5 won't come before 2024 either.

Hardly an issue if Arrowlake, a much better product comes in "almost at the same time" and be labelled the same generation. I can expect everything above xx600K to be Arrowlake for example. Actually considering 13700K is 8+8, it can replace even that. But since gamers are very finicky bunch, better they have Arrowlake just for K chips.

Well the problem is that if Arrowlake being a 2024 product with 8P cores means that Intel does not have a good answer against Zen 4 3D V-Cache next year. Like I don't see the 13900KS being a good solution against V-Cache CPUs who'll be power sipping in comparison.