Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads

[Tigerick]

Wildcat Lake (WCL) Specs

Intel Wildcat Lake (WCL) is upcoming mobile SoC replacing Raptor Lake-U. 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 Q1 2026.

		Intel Raptor Lake U	Intel Wildcat Lake 15W?	Intel Lunar Lake		Intel Panther Lake 4+0+4
Launch Date		Q1-2024	Q2-2026	Q3-2024		Q1-2026
Model		Intel 150U	Intel Core 7	Core Ultra 7 268V		Core Ultra 7 365
Dies		2	2	2		3
Node		Intel 7 + ?	Intel 18-A + TSMC N6	TSMC N3B + N6		Intel 18-A + Intel 3 + TSMC N6
CPU		2 P-core + 8 E-cores	2 P-core + 4 LP E-cores	4 P-core + 4 LP E-cores		4 P-core + 4 LP E-cores
Threads		12	6	8		8
Max Clock		5.4 GHz	?	5 GHz		4.8 GHz
L3 Cache		12 MB		12 MB		12 MB
TDP		15 - 55 W	15 W ?	17 - 37 W		25 - 55 W
Memory		128-bit LPDDR5-5200	64-bit LPDDR5	128-bit LPDDR5x-8533		128-bit LPDDR5x-7467
Size		96 GB		32 GB		128 GB
Bandwidth				136 GB/s
GPU		Intel Graphics	Intel Graphics	Arc 140V		Intel Graphics
RT		No	No	YES		YES
EU / Xe		96 EU	2 Xe	8 Xe		4 Xe
Max Clock		1.3 GHz	?	2 GHz		2.5 GHz
NPU		GNA 3.0	18 TOPS	48 TOPS		49 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.

 

[adroc_thurston]

They do suck but far and away from 15-20W ST stuff.

Eh, 8-10W 1t isn't that far off anyway.
(Remember that ARM wanted 3.5GHz X1 for laptops where phones toasted themselves at like 3)

X cores are a bit bizarre now with the L1I and width decisions, they will be turkey basted by Nuvia if things even half pan out.

Yea X4 is funny as hell, 10 decode, 10 goddamned ALUs what even.

+ Apple meltdowns are going to be be pretty A1.

I genuinely wanna see mactoddlers cope; they've been a wee bit too obnoxious in the recent past.

What's your forecast on LNC IPC gain?

>25% IPC with a wee bit of clk regression and they've clawed back some area from what I know of ARL/LNL.

I think Intel's E Cores will still be alright if only because I don't see any other Arm competition (and Nvidia stuff is niche for HPC),

The issue is SRF being relatively puny core count so it's just a mid chip that's upfront cheap targeting the market where upfront costs matter relatively little.

but if Zen 5 shapes up like it should then AMD will win the next generation too for cloud providers.

3 out of s8 already dueled as to who gets the first dibs on Turin-D so it's good.

 

[jpiniero]

I start wondering who sent these confidential materials to MLID......

As if MLID didn't just make it himself.

 

[SpudLobby]

Eh, 8-10W 1t isn't that far off anyway.
(Remember that ARM wanted 3.5GHz X1 for laptops where phones toasted themselves at like 3)

Yea X4 is funny as hell, 10 decode, 10 goddamned ALUs what even.

I genuinely wanna see mactoddlers cope; they've been a wee bit too obnoxious in the recent past.

>25% IPC with a wee bit of clk regression and they've clawed back some area from what I know of ARL/LNL.

The issue is SRF being relatively puny core count so it's just a mid chip that's upfront cheap targeting the market where upfront costs matter relatively little.

3 out of s8 already dueled as to who gets the first dibs on Turin-D so it's good.

Eh, 8-10W 1t isn't that far off anyway.
(Remember that ARM wanted 3.5GHz X1 for laptops where phones toasted themselves at like 3).

Yes I remember that though I don't know 3.5 would've hit that but maybe on Samsung yeah with an X2. I suspect X4's on a laptop now at 3.5-3.7GHz would just be 6-7W which is fine it's just a weird core. 12-20+ [ST] is still aggressively retarded to me in a laptop, but to each his own.

I genuinely wanna see mactoddlers cope; they've been a wee bit too obnoxious in the recent past.

Ha. I reluctantly use one pal, much like Exist50 I assume. Many cases currently, because some don't wants to get bent over by sloppiness in the Windows camp. I do trust QC to get it as long as there's noting massively egregious going on (e.g. dvfs rumor) - they obviously do idle power, scheduling, just fine from mobile. if Nuvia is just M1-tier on perf and efficiency with decent product availability I'll switch back and maybe grab a Z5 APU for a desktop too depending on how that goes with RDNA's emergency plastic surgery in 3.5.


Dunno what Z5 timeframe really looks like though for mobile or desktop other than CES intro for one.

 

[SpudLobby]

I don't want to belabor the point, but I do think you're just mistaking them moving that die around the roadmap. But if it's canceled, we would hopefully hear something during their earnings report.

Well, LNL will primarily be in laptops. It's an Apple M-series competitor. I've been saying for a while now that AMD really needs a fanless ~10W part. They have all the IP needed. Just need to get the platform priories in order.

FWIW I think worth noting when we say "10W" since this throws some off:

You could still use this stuff in a device with a fan. It's not really Atom Goldmont-tier budget stuff, and there's plenty of performance to go around. It won't be as efficient as something ARL 6+8 at say a CB23 MT run pegged to 25W (imagine same node for a moment, more cores = lower voltages, so at some point it'll be more efficient on an MT curve vs LNL taxed out), but that's not the point of course.

For ex, an M2 still gets pretty close to a 7840U in GB6, 10% off despite 4+4 over 8. This stuff is pretty performant if needed and I expect Intel's to be the same, so we'll see how LNL is. AMD definitely ought to give a -profanity removed- about mobile in the more broad sense of sub-25/28W. Things that *can* go fanless but would virtually never throttle with really basic fans, have low idle draw etc.

Profanity is not allowed in the tech forums.

Daveybrat
AT Moderator

 

[SpudLobby]

Those numbers are a little off, GNR will not be 160Mtr/mm2 when you account for the whole die. Intel 3 would only hit 160Mtr/mm2 on a theoretical smaller compute tile like MTL for example.

The TSMC numbers (per angstronomics)
N5 HP Cell: 92.3 MT/mm²
N4 HP Cell: 97.8 MT/mm²
N3B HP Cell: 124.00 MT/mm² (N3E is slightly less)

This is only comparing HP cells though. The N3 HD cells probably have better density, leakage and yield that could give an advantage to low power devices compared to Intel 3.

Would really like to see a good Intel 3 HD part, wish that's what Lunar Lake was if it were competitive.

 

[SpudLobby]

On the topic of the merits of SMT, I think the big question is what they're going to do for servers. GNR is RWC, so that's fine, and they haven't talked about a big core server product (Diamond Rapids?) despite naming Clearwater Forest in 2025, so it's possible we don't see DMR till 2026. That gives them 1-2 gens to either add back SMT (either for everything, or a dedicated server fork), or find some replacement solution. Because the loss of SMT, all else equal, will definitely hurt them in throughput efficiency (both area and power), plus with any CSP renting out virtual cores (fewer units per chip).

For client, now that there's a bunch of E cores, I think things are fine either way. Don't need a million threads for client chips.

Agreed, servers is obviously where it hurts though I was under the impression they might preserve it in LNC for servers. On the other hand their E cores are ridiculously small (Look at Crestmont) and if they keep most of AVX-512 off and the uArch smaller, I could see their E Core stuff still selling well since HT isn't a big deal there (and they surely know that). Might have a slight price competitive advantage (albeit at lower perf) over AMD if Intel are able to ram similar cores in less area, I don't know, but there's not really any formidable Arm server competition that isn't just:

A). Hyperscalers like Graviton3 - nothing to laugh at in light of their power efficiency and density but also not widely available everywhere. Growing but Amazon-exclusive. MS and Google remain to be seen with a Neoverse implementation or otherwise.
B). Nvidia - niche, Grace is just an HPC bundle with H100's, too expensive and nnot broad.
C). Ampere - Ampere sucks.

There's really not even a rising analogue to Qc/Nuvia in laptops like Qualcomm of Arm cores soon to release a server part - I mean speaking of them literally, they actually were going to, and then canceled it which was smart, but yeah.

So it's really AMD, Intel on merchant general purpose server CPU's. I don't think Intel is totally out of it though.

 

[adroc_thurston]

AMD definitely ought to give a Redacted about mobile in the more broad sense of sub-25/28W.

They've been the 15W champs for a while already.

I could see their E Core stuff still selling well since HT isn't a big deal there

They need a lot, lot more atoms to justify their existence then.
SRF ain't cutting it.

 

[SpudLobby]

They need a lot, lot more atoms to justify their existence then.
SRF ain't cutting it.

We'll see. If Zen 5c if it follows the same design and arch pattern as 4c did to 4 (and I assume it might, and AMD wants to core count max with 3NM 5c for servers), it gets dicier for Intel unless they have something up their sleeve for future Mr. Monts.

Either way I suspect even when/if they "rebound" as rabid fans are into, that just amounts to being competitive whatsoever, e.g. they're taking a a revenue or margins cut to some degree, the monopoly is never going to return. Competition isn't going anywhere, and the 00's, 10's aren't coming back.

 

[TESKATLIPOKA]

I am interested how good meteor Lake will end up compared to Raptor.
From the leaks the 15W segment will see an increase in P-cores. Now the question is at what clocks will 4P+8E work at 15W.
i7 1365U with 2P+8E has base clocks 1.8GHz and 1.3GHz at 15W.

On the other hand, I expect a significant uplift with the IGP.

 

[adroc_thurston]

If Zen 5c if it follows the same design and arch pattern as 4c did to 4

Compacts are always the exact same core, just with a low fmax floorplan.

Either way I suspect even when/if they "rebound" as rabid fans are into

Intel looks fine in client, it's server where their prospects are really-really grim.

Now the question is at what clocks will 4P+8E work at 15W.

Pretty bad.
MTL doesn't clock all that well, really.

 

[SpudLobby]

Compacts are always the exact same core, just with a low fmax floorplan.

Yeah figured. It's a fantastic strategy in AMD's case, just had no clue if they might change it.

Intel looks fine in client, it's server where their prospects are really-really grim.

Surprised we agree. Pretty much how I feel looking at their next 2-3 years. DC execution (or lack thereof) is just painful to watch.

 

[SpudLobby]

MTL clocks on PL1 look alright though no? It's just PL2 at peak, no change. That still matters of course and sucks, but setting to a balanced power plan I think if those PL1 rumors are accurate to the actual power draw and frequency, it's a significant improvement for some cases. Not what it could be though.

 

[adroc_thurston]

just had no clue if they might change it.

Ah, no, it's a quick and dirty way to get a cloud-native(tm) core and in context of Zen5(c)-onwards it truly works wonders.

DC execution (or lack thereof) is just painful to watch.

Yea, man, grim times ahead.
At least the GPGPU arena seems more competitive than ever.

MTL clocks on PL1 look alright though no?

None of those chips sit at PL1 rated clock, like, ever so yea.

 

[SpudLobby]

Ah, no, it's a quick and dirty way to get a cloud-native(tm) core and in context of Zen5(c)-onwards it truly works wonders.

Obviously.

Yea, man, grim times ahead.
At least the GPGPU arena seems more competitive than ever.

Ya.

None of those chips sit at PL1 rated clock, like, ever so yea.

Depends of course. At least if you wanted to change the power policy via vendor or the Windows panel, you could reap the benefit. Yes, out of the box maxxed out it’ll guzzle.

 

[gdansk]

I am interested how good meteor Lake will end up compared to Raptor.
From the leaks the 15W segment will see an increase in P-cores. Now the question is at what clocks will 4P+8E work at 15W.
i7 1365U with 2P+8E has base clocks 1.8GHz and 1.3GHz at 15W.

On the other hand, I expect a significant uplift with the IGP.

Since the 780M was a bit of a dud, is it likely this eclipses it in general game performance?

 

[DAPUNISHER]

Since the 780M was a bit of a dud, is it likely this eclipses it in general game performance?

I suspect the drivers will hold it back regardless of whatever potential the hardware has. I hope I'm wrong.

 

[TESKATLIPOKA]

Since the 780M was a bit of a dud, is it likely this eclipses it in general game performance?

It has 1/3 more EU, but we don't know the final clockspeed, but at higher TDP I expect a significant increase.
Then the drivers is also a problem as @DAPUNISHER already said.

 

[Khato]

Since the 780M was a bit of a dud, is it likely this eclipses it in general game performance?

Unlikely? But it should definitely be competitive. We can already compare the 780M against the A350M and A370M and MTL graphics is likely to fall somewhere between those two.

 

[mikk]

It has 1/3 more EU, but we don't know the final clockspeed, but at higher TDP I expect a significant increase.
Then the drivers is also a problem as @DAPUNISHER already said.

There are several MTL entries and some of them have a GPU frequency of 2.25 Ghz. I think this is roughly the expected clock speed for the faster models.

MTL GPU benefits from the Alchemist driver work, actually the driver should be in a better shape than the old Xe LP. Intel is clearly more focused in Alchemist.

 

[TESKATLIPOKA]

There are several MTL entries and some of them have a GPU frequency of 2.25 Ghz. I think this is roughly the expected clock speed for the faster models.

MTL GPU benefits from the Alchemist driver work, actually the driver should be in a better shape than the old Xe LP. Intel is clearly more focused in Alchemist.

Very nice, that's 50% higher boost. I knew Alchemist could clock at >2GHz, but I wasn't sure what we could expect with limited TDP and higher EU count.

45W Raptor Lake-H 96EU IGP clocks up to 1.5GHz.
28W Raptor Lake-P 96EU IGP clocks up to 1.5GHz.
15W Raptor Lake-U 96EU IGP clocks up to 1.3GHz.

Meteor should look like this then:
45W Meteor Lake-H 128EU IGP clocks up to 2.25GHz.
28W Meteor Lake-P 128EU IGP clocks up to 2.25GHz.
15W Meteor Lake-U 128EU IGP clocks up to 1.95GHz.

It's yet to be seen If the average clockspeed will be 50% higher at limited TDP, but It looks very good. We can roughly expect 2x better TFLOPs.

TimeSpy Graphics:
i7-1280P 96EU IGP 1450Mhz -> 1871 points (100%)
Beelink GTR7 7840HS -> 2910 points (156%)
A370M 128EU 1550Mhz + GDDR6 -> 3899 points (208%)

From this, you would guess Meteor Lake IGP should be much faster than A370M, which has the same specs, but lower clocks.
We don't know what was the average clockspeed for 1280P IGP or what will be the average clockspeed for Meteor Lake and A370M had 14Ghz GDDR6, so It could perform worse than A370M.

On the other hand, It's quite possible that Meteor Lake IGP will end up faster than 780M, If the game is optimized for Intel.

 

[JoeRambo]

Intel is making smart move with removing HT from big client cores. In fact i have been disabling HT since i have got 10C Comet Lake and never looked back. 8 cores of GLC is plenty for anything on workstation that is not doing compiles non stop. I have used to add "rendering" to this list, but nowadays it is done on GPUs.
HT is not free, even if we ignore costs that are mostly relevant for Intel's beancounters like design, validation etc, we end up with probably 5-10% area cost. And all HT stuff is in very hot path like register files, L1, OoO structures and buffers, so it is non zero cost in clocks either.
I also think that these costs are probably growing big time due to security requirements -> adversarial thread running on same core can probably leak out enough info about other SMT thread. ( that means that in cloud setting care needs to be taken to sell both SMT thread to same customer ).

So reality is that proper heavy workloads ( think yCruncher, all those uber slow well optimized software video/image encoders etc ) are not gaining anything or even loosing perf and those who benefit from additional threads are the ones that can make great use of E cores.
Apple has shown that having balanced memory subsystem and large OoE engine serving one very powerful thread is good both for performance and power efficiency. And that using HT to extract last 10-20% of performance from badly balanced and underfed core is not that great idea.

So on client future seems to be very powerful P-Cores backed by as many E cores as marketing segment desires and maybe even backed by few more energy optimized E-Cores in low power SoC area.
On servers HT would still make sense as there are plenty of workloads that are inherently benefitting from HT like databases, JVM, VMs etc.

 

[DrMrLordX]

And that using HT to extract last 10-20% of performance from badly balanced and underfed core is not that great idea.

HT has historically helped Intel run poorly-optimized MT code well when older AMD architectures (notably anything from the CON core lineup) bungled the task severely. Look at Dr. Cutress' 3DPM v1. HT on vs. off was nearly a 50% performance gain for HT on.

 

[A///]

I'm missing a lot of conversation and don't feel like reading. hopefully gedda sees this post and is nice enough to give me a summary since he's usually very happy to.

that put forth I hate to repeat myself like a crazy person but I've been saying forever that Intel's client will always be afloat due to how much they can supply vs amd. their outstanding dc contracts are helping but time is ticking. dc will always have better margins than anything else and amd is eating their lunch. I haven't done much digging into the new xeon max processors foronicks tested but the max with hbm seem to eat at the 9654, but if I'm not mistaken the mkl library is being access here although I thought AMD found a workaround for that. either way the max likely use far more power, it is intel of course not a second return of jesus. if amd mess up their tr 7000 launch later in a few months intel may have the upper hand even if w3400 is a pile of crap.

 

[adroc_thurston]

but I've been saying forever that Intel's client will always be afloat due to how much they can supply vs amd.

That's not the real reason, PC volumes have cratered a while ago.

the max with hbm seem to eat at the 9654

And in the end gets eaten by Genoa-X clusters.

 

[eek2121]

I am interested how good meteor Lake will end up compared to Raptor.
From the leaks the 15W segment will see an increase in P-cores. Now the question is at what clocks will 4P+8E work at 15W.
i7 1365U with 2P+8E has base clocks 1.8GHz and 1.3GHz at 15W.

On the other hand, I expect a significant uplift with the IGP.

CPU performance uplift will mostly be multicore. Perf/watt much improved.

Compacts are always the exact same core, just with a low fmax floorplan.

Intel looks fine in client, it's server where their prospects are really-really grim.

Pretty bad.
MTL doesn't clock all that well, really.

3.4 ghz base is a big improvement from Raptor Lake. That is all-cores mind you.

Peak clocks won’t regress much, if at all over raptor lake. The last I saw these chips would have a peak of around 4.8-4.9ghz, but that has probably changed.

Finally (to no one in particular), what is up with you folks throwing peak per core power numbers around? That number is kind of meaningless. A single core on a Zen 4 chip may consume 20W, but you will never see it consume 20W in an all-core workload.

A well-designed CPU should be able to scale as much as possible as long as power limits haven’t been reached. If the power limit is 28W there is absolutely nothing wrong with a single core using nearly all of it.

Also, race to idle is a thing. As long as chip a can complete tasks quicker and use less total power than chip b, chip a is the winner, even if chip a peaks at 28W per core vs chip B’s 5-10W per core.