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

[Tigerick]

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.



Comparison of upcoming Intel's U-series CPU: Core Ultra 100U, Lunar Lake and Panther Lake

Model	Code-Name	Date	TDP	Node	Tiles	Main Tile	CPU	LP E-Core	LLC	GPU	Xe-cores
Core Ultra 100U	Meteor Lake	Q4 2023	15 - 57 W	Intel 4 + N5 + N6	4	tCPU	2P + 8E	2	12 MB	Intel Graphics	4
?	Lunar Lake	Q4 2024	17 - 30 W	N3B + N6	2	CPU + GPU & IMC	4P + 4E	0	12 MB	Arc	8
?	Panther Lake	Q1 2026 ?	?	Intel 18A + N3E	3	CPU + MC	4P + 8E	4	?	Arc	12

Comparison of die size of Each Tile of Meteor Lake, Arrow Lake, Lunar Lake and Panther Lake

	Meteor Lake	Arrow Lake (20A)	Arrow Lake (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	24 MB ?	36 MB ?	12 MB	?
tCPU	66.48
tGPU	44.45
SoC	96.77
IOE	44.45
Total	252.15

Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[511]

Arctic wolf will be 12 wide.. so another big jump

I hope it is another 40% it would match M3 P core IPC in 2-3 years

 

[DavidC1]

Skymont being exactly 38% more IPC is pretty impressive. No wonder why Intel is looking to consolidate the core team, the Monts could overtake the Coves. Perhaps as early as Darkmont? Or maybe it’ll be the Wolves that get it done.

I would have thought with straight up doubling of FP units, the 68% number would have applied to Cinebench, or at least somewhere in between 38-68%. 38% over LP-E, and if you divide by 5% you get 31%, which is the ~30% number Intel has been getting every generation since Goldmont.

I hope it is another 40% it would match M3 P core IPC in 2-3 years

They've been doing 30% uarch and 10-15% clock gains. I had some feeling they made up for contemporary competing ARM Cortex architecture with somewhat higher clocks.

My hope is that they'll cut the pipeline stages to 11-12, add a large 128KB L1-I and 64KB L1-D cache for power efficiency, aim for 50% gain per clock but 10% lower clocks.

Perhaps as early as Darkmont?

Darkmont seemed like a Crestmont like change. Though Raichu talked about front-end changes. 128KB L1-I for power efficiency? Please?

 

[511]

I would have thought with straight up doubling of FP units, the 68% number would have applied to Cinebench, or at least somewhere in between 38-68%. 38% over LP-E, and if you divide by 5% you get 31%, which is the ~30% number Intel has been getting every generation since Goldmont.

Ring and no ring also makes a difference so we will have 2 see in ARL Review

Darkmont seemed like a Crestmont like change. Though Raichu talked about front-end changes. 128KB L1-I for power efficiency? Please?

5-6% ipc over this is basically within 15% of Zen 5 and 5c is approx 96% zen 5 ipc so within 10% of darkmont so basically 288 Non HT core Clearwater forest is just gonna be a monster workstation

 

[jdubs03]

5-6% ipc over this is basically within 15% of Zen 5 and 5c is approx 96% zen 5 ipc so within 10% of darkmont so basically 288 Non HT core Clearwater forest is just gonna be a monster workstation

Is that including Skymont being equal to 2% more than Raptor Cove (iso-frequency, which I presume was 4.6Ghz)? Seems like it’d be within 12% of Lion Cove, which should be around/higher than Zen 5. So that 5-6% gain would get it very close to Zen 5c.

 

[511]

Is that including Skymont being equal to 2% more than Raptor Cove (iso-frequency, which I presume was 4.6Ghz)? Seems like it’d be within 12% of Lion Cove, which should be around/higher than Zen 5. So that 5-6% gain would get it very close to Zen 5c.

my bad I took the 21% diffrence between LNC and SKT based on this

What I heard is on ARL, lion cove is 442 per core/ghz , Skymont is 363 per core/ghz, in the r23.

So i accidentally made my calculations on this one assumig LNC ~ Zen 5 but zen 5 is few % slower anyway it would be shrinked to 5% per core performance but with 288 cores/thread vs 192 cores/thread

 

[SiliconFly]

Arctic wolf will be 12 wide.. so another big jump

12 wide doesn't automatically mean higher IPC. Also, any idea who first mentioned Arctic Wolf is 12 wide?

Plus arrow refresh will have higher clocks. 6ghz on n3b mayb ? 🤔

I don't think TSMC has Intel 7 UHP equivalent libraries that clocks real high. Even if they have, I don't think Intel will use them anymore due to efficiency reasons.

IIRC the whole ARL lineup's cost is almost 2 times higher than Zen5 including Strixpoint. Though Intel has larger market share to mitigate this.

I don't think it'll be anywhere near 2X the cost. Maybe a bit higher. When a company has massive volume like Intel, Economies of Scale kicks in to reduce cost. Thats one of the primary reasons why they're using tiles & foveros despite costs.

Sierra Forest yield is not good, at least it's worse than expected.

It's a false rumor. It has already been addressed by other people in this thread.

Skymont being exactly 38% more IPC is pretty impressive. No wonder why Intel is looking to consolidate the core team, the Monts could overtake the Coves. Perhaps as early as Darkmont? Or maybe it’ll be the Wolves that get it done.

Darkmont is supposed to be a minor incremental upgrade. My guess is, it might just be the (almost) same Skymont but on 18A.

 

[511]

12 wide doesn't automatically mean higher IPC. Also, any idea who first mentioned Arctic Wolf is 12 wide?


I don't think TSMC has Intel 7 UHP equivalent libraries that clocks real high. Even if they have, I don't think Intel will use them anymore due to efficiency reasons.


I don't think it'll be anywhere near 2X the cost. Maybe a bit higher. When a company has massive volume like Intel, Economies of Scale kicks in to reduce cost. Thats one of the primary reasons why they're using tiles & foveros despite costs.


It's a false rumor. It has already been addressed by other people in this thread.


Dark mont is supposed to be a minor incremental upgrade.

Darkmont is on 18A which is in between N3P and N2 so 4-5% ipc + 10-15% ppw from node means it will be 15% Better iso core performance/watt

 

[SiliconFly]

Darkmont is on 18A which is in between N3P and N2 so 4-5% ipc + 10-15% ppw from node means it will be 15% iso core performance

Switching to a better node doesn't automatically increase IPC for a given processor. But, it can be clocked higher at the same power provided the node has higher Fmax. Another advantage is, a better node offers higher densities there by offering higher transistor budget for improving certain structures like caches without any major redesign thereby increasing performance. Also, any architectural upgrades may provide additional performance boost.

Otherwise, no. So, just putting Skymont on 18A or 14A or 10A ... and calling it Darkmont will not have any additional IPC. And will not have any extra performance for a given clock.

 

[511]

I missed performance per core per watt 😅

 

[Henry swagger]

Switching to a better node doesn't automatically increase IPC for a given processor. But, it can be clocked higher at the same power provided the node has higher Fmax. Another advantage is, a better node offers higher densities there by offering higher transistor budget for improving certain structures like caches without any major redesign thereby increasing performance. Also, any architectural upgrades may provide additional performance boost.

Otherwise, no. So, just putting Skymont on 18A or 14A or 10A ... and calling it Darkmont will not have any additional IPC. And will not have any extra performance for a given clock.

I think darkmont will be like crestmomt 5% ipc or 10 anf arctic wilf will be the big jump

 

[FlameTail]

Arctic wolf will be 12 wide.. so another big jump

Wow. X86 cores surpassing ARM cores in width. Didn't expect this to happen so soon.

 

[SiliconFly]

I think darkmont will be like crestmomt 5% ipc or 10 anf arctic wilf will be the big jump

Thats the general consensus (and I think you meant Skymont).

Wow. X86 cores surpassing ARM cores in width. Didn't expect this to happen so soon.

I think we're just speculating here. No concrete info yet I believe.

 

[511]

The thing is 3*3 decode is cheaper than 8 wide decode in area

 

[igor_kavinski]

The thing is 3*3 decode is cheaper than 8 wide decode in area

What's the catch? How about 6*6?

 

[FlameTail]

The thing is 3*3 decode is cheaper than 8 wide decode in area

Is it?

Edit: As you can see, the x86 decoder is much more complex.

 

[FlameTail]

What laptop price segment would Wildcat Lake belong to?

 

[SiliconFly]

Is it?

Yep. A 3x3 decoder is simpler than a 8 wide decoder.

 

[Abwx]

My prediction for Lion Cove based on leaks seen so far is ~6-8% average IPC over Zen 5.

As usual it wil be only on Cinebench, you can bet that with any other renderer Zen 5 will be ahead like Zen 4 wich was behind RPL in CB and ahead for all other renderers.

 

[511]

As usual it wil be only on Cinebench, you can bet that with any other renderer Zen 5 will be ahead like Zen 4 wich was behind RPL in CB and ahead for all other renderers.

In we only count absolute single core than no

 

[naukkis]

Yep. A 3x3 decoder is simpler than a 8 wide decoder.

For x86 which is decoding nightmare. To Arm it does not comply.

 

[jdubs03]

On single thread, the 285K will be ahead. Looking at computerbase, the 14900K was 9% ahead of the 7950X. I don’t think it’ll be ahead by nearly that much. I said 0-3% before, I think it’ll be around that 3%. In multithread the 7950X was level with the 14900K. I think the 285K will at least be equal again.

 

[igor_kavinski]

I think the 285K will at least be equal again.

This is Intel's what? Fourth CPU gen after they got smacked in the face with V-cache? I hate that they don't care about going all in with a large cache, at least on their flagship SKU. It just shows that they are confident that their lame userbase will gulp up whatever they throw at them.

 

[Abwx]

In we only count absolute single core than no

Even in single core, we ll see in a month when they release ARL but given the gap between the 9950X and the 14900K i doubt that it can be closed given the improvement in CB, as with other renderers it will start from a lower step.

FTR the former is 21% faster in Blender, 30% faster in Corona, 51% faster in V-Ray and only 5% faster in CB 2024, wich by those number is the convenient exception used as a rule, the geomean average of those 4 numbers, 25.6%, is way more representative than an outlier.

As you can see it require a much better ST perf to get those numbers given that SMT bring a comparable gain for all tests.

 

[cebri1]

Even in single core, we ll see in a month when they release ARL but given the gap between the 9950X and the 14900K i doubt that ican be closed given the improvement in CB, as with other renderers it will start from a lower step.

FTR the former is 21% faster in Blender, 30% faster in Corona, 51% faster in V-Ray and only 5% faster in CB 2024n wich by those number is the convenient exception used as a rule, the geomean average of those 4 numbers, 25.6%, is way more representative than an outlier, and is the gap to close.

Talk about cherry picking

 

[igor_kavinski]

Even in single core, we ll see in a month when they release ARL but given the gap between the 9950X and the 14900K i doubt that ican be closed given the improvement in CB, as with other renderers it will start from a lower step.

Intel's benefit this time is that they can count on real physical cores and not be hampered by HT related issues dragging down performance. I think you should brace yourself for 9950X losing in some benchmarks because SMT cannot match a real core. I just have to call it like it is. Intel has a slight advantage this time around. Well, until AMD boosts 9950X performance with an AGESA update or something. I would've preferred 9950X having CCD2 comprised of 16 Zen5c cores.