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	8 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)	Arrow Lake Refresh (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Desktop Only	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2025 ?	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	8P + 32E	4P + 4E	4P + 8E
LLC	24 MB	24 MB ?	36 MB ?	?	8 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.)

 

[Gzxy]

Which part of the system is LionCove because I can't recognize it?

left 4 Skymont, right 4 Lion Cove. Below Skymont is it's cache iirc and between the 4 Lion is their cache.

 

[Ghostsonplanets]

Which part of the system is LionCove because I can't recognize it?

View attachment 97425
left 4 Skymont, right 4 Lion Cove. Below Skymont is it's cache iirc and between the 4 Lion is their cache.

Also, bit of a guessing from Olrak:

https://twitter.com/x/status/1778297701576438265

"ThunderShock":

Although it is just a rendering instead of the floorplan, it still has some questions on label.
The NPU site looks like the ME, and I think VPU/NPU should be close to the IMC, the above of ME.🤔

https://twitter.com/x/status/1778433229856399849

"If this is an accurate depiction of the floorplan, that would mean a Skymont cluster is now around 2 Lion Cove's worth of Area

I can see why people are hyping SKT as the best part of ARL. They seem to be heavily investing area into E-cores"

 

[mikk]

Core size numbers from here. Skymont grows big time unlike Lion Cove versus MTL on Intel 4.

https://twitter.com/x/status/1778860445853491346

Lunar Lake:
P core: 4.55mm²
E core cluster: 8,1mm²
E core (without L2): 1.52mm²

MTL:
P: 5,33mm²
E cluster: 5,9mm²
E core only: 1,04mm²

ADL:
P: 7,12mm²
E cluster: 8,3mm²
E core only: 1,59mm²

 

[Ghostsonplanets]

Core size numbers from here. Skymont grows big time unlike Lion Cove versus MTL on Intel 4.

https://twitter.com/x/status/1778860445853491346

Lunar Lake:
P core: 4.55mm²
E core cluster: 8,1mm²
E core (without L2): 1.52mm²

MTL:
P: 5,33mm²
E cluster: 5,9mm²
E core only: 1,04mm²

ADL:
P: 7,12mm²
E cluster: 8,3mm²
E core only: 1,59mm²

Intel must be taking a lot of advantage from TSMC N3 density and perhaps even some denser libraries instead of the usual UHP or HP libraries they use. I can't remember the last time an Intel Core was so small.

And Lion Cove is supposed to be an even wider and deeper core. Quite interesting.

 

[AMDK11]

View attachment 97425
left 4 Skymont, right 4 Lion Cove. Below Skymont is it's cache iirc and between the 4 Lion is their cache.

The question now is whether the structure of Lion Cove matches the actual system. Is it just loose graphics without maintaining the correct structure with image confirmation?

LionCove has 2.5-3MB L2.

EDIT:
This is the same level of visualization as RedwoodCove in Meteorlake, which does not have much to do with the actual structures of the P core.
I'm still waiting for an actual photo of the LionCove core structure.

 

[Ghostsonplanets]

Doesn't Lion Cove also introduce a new L0 cache? Or I'm dreaming?

 

[Gzxy]

Intel must be taking a lot of advantage from TSMC N3 density and perhaps even some denser libraries instead of the usual UHP or HP libraries they use. I can't remember the last time an Intel Core was so small.

And Lion Cove is supposed to be an even wider and deeper core. Quite interesting.

P-cores dropped hyper threading. That thing takes lots of space.

P-cores are clocked probably a lot lower than previous gens for efficiency and thus they use as you said high density libs. I am not sure but i believe all previous gens U, H, HX P-cores where the same exactly and they just modified clocks and disabled parts of the cache. This is not the case since LNL is not meant for high TDPs.

https://twitter.com/x/status/1766693553080807723

 

[Ghostsonplanets]

P-cores dropped hyper threading. That thing takes lots of space.

Oh! I wasn't aware HT was that much of an area eater

P-cores are clocked probably a lot lower than previous gens for efficiency and thus they use as you said high density libs

Xino shared that LNL EV-QS can reach 4.9GHz max turbo clock, so it isn’t that low clocked. Still, 4.9GHz would probably be feasibly with HD libs given how AMD was able to clock in the same range with Zen 2/Zen 3.

 

[SiliconFly]

If a "Skymont cluster is now around 2 Lion Cove's worth of Area", then I think the chances of us seeing ARL 8P+32E in pretty much non-existent!

Makes the die way too large and very expensive at the cost of serious yield issues. Looks grim.

So, there's this guy... they call him dimlid; they say he can predict the future...

 

[Hulk]

Raptor Cove has about 45% or so better IPC than Gracemont when it comes to single core. So while the increased size of Skymont is enticing I find it hard to believe Skymont will approach Raptor Cove IPC, which would be needed for Arrow Lake to compete with Raptor Lake in MT.

Houston, we have a problem.

 

[Saylick]

Oh! I wasn't aware HT was that much of an area eater

It shouldn’t be. Iirc, SMT only added like 10% more core area than without it. Additionally, we aren’t sure if LNC has SMT but it’s just turned off. I suspect it wasn’t designed from the ground up to be a pure non-SMT design, but the disabled SMT allows for easier validation, which means faster time-to-market, and less chance of security vulnerabilities.

Edit: spelling.

 

[Hulk]

It shouldn’t be. Iirc, SMT only added like 10% more core area than without it. Additionally, we aren’t sure if LNC has SMT but it’s just turned off. I suspect it wasn’t designed from the ground up to be a pure non-SMT design, but the disabled SMT allows for easier validation, which means faster time-to-market, and less chance of security vultures.

Also while the lack of HT will diminish MT performance among the P cores, it will allow the P's to reach higher clocks with less power and heat when there are compute intensive threads thrown their way. Of course this assumes the E's are capable of handling all of the "lighter work" the missing logical cores would have been working on.

 

[Markfw]

Also while the lack of HT will diminish MT performance among the P cores, it will allow the P's to reach higher clocks with less power and heat when there are compute intensive threads thrown their way. Of course this assumes the E's are capable of handling all of the "lighter work" the missing logical cores would have been working on.

I think the reason why Intel dropped HT was, that their implementation was far more susceptible to security holes than AMDs, and this was an easy way to eliminate that as well as higher clock speeds and a loss of only max 20% performance, and allowed the e-cores to make it up ??

 

[Henry swagger]

Raptor Cove has about 45% or so better IPC than Gracemont when it comes to single core. So while the increased size of Skymont is enticing I find it hard to believe Skymont will approach Raptor Cove IPC, which would be needed for Arrow Lake to compete with Raptor Lake in MT.

Houston, we have a problem.

It will.. wait for benchmark leaks

 

[coercitiv]

Also while the lack of HT will diminish MT performance among the P cores, it will allow the P's to reach higher clocks with less power and heat when there are compute intensive threads thrown their way.

Feels like we've been over this before, removing SMT is an efficiency loss, not a win. The catch is the loss will only exist in benchmarks, in most consumer workloads the number of threads available from the P+E combo will be high enough to out-scale the software. As for the area improvements, I would really wait and see for the benefits of removing SMT and AVX-512. There is a non-zero chance these features are just disabled in ARL.

 

[ondma]

Feels like we've been over this before, removing SMT is an efficiency loss, not a win. The catch is the loss will only exist in benchmarks, in most consumer workloads the number of threads available from the P+E combo will be high enough to out-scale the software. As for the area improvements, I would really wait and see for the benefits of removing SMT and AVX-512. There is a non-zero chance these features are just disabled in ARL.

I have mentioned this before, but it makes one wonder how loss of HT will affect gaming. Without E cores, and with lower core counts, we saw a definite lack non HT cpus to handle higher gaming workloads (e.g. 2500k vs 2600K). Did HT help gaming with Raptor Lake, or do the E cores step in? Is eight strong P cores by themselves enough?

 

[DrMrLordX]

Is eight strong P cores by themselves enough?

People seemed to be okay with 8P cores on Raptor Lake when playing most games, though none of them appeared to have disabled HT and e-cores at the same time (usually just e-cores).

 

[coercitiv]

I have mentioned this before, but it makes one wonder how loss of HT will affect gaming. Without E cores, and with lower core counts, we saw a definite lack non HT cpus to handle higher gaming workloads (e.g. 2500k vs 2600K). Did HT help gaming with Raptor Lake, or do the E cores step in? Is eight strong P cores by themselves enough?

With games we are still at a point where a strong 6-core is better than 8+ weaker cores. From 6 cores onward the memory subsystem matters more (cache, memory speed etc). Look at how close 14600K gets to 14700K, that's probably just the difference in L3 cache. We could also look at 7600X vs. 7700X, there we have influence from E cores and the L3 cache is identical.

Spoiler: 14600k & 7600X gaming perf

I would argue that 8C/8T will do fine in the next few years with the support of E cores and maybe some more APO style optimizations. Cache size, speed and overall latencies (memory, core-to-core etc) will matter a lot more than the presence of SMT.

 

[igor_kavinski]

I suspect it wasn’t designed from the ground up to be a pure non-SMT design, but the disabled SMT allows for easier validation, which means faster time-to-market, and less chance of security vulnerabilities.

I hope you are right. This could open the possibility of an SMT enablement firmware patch post launch, maybe months or even a year later.

 

[igor_kavinski]

Sorry wrong thread.

 

[igor_kavinski]

Sorry. Wrong thread.

 

[Hulk]

Feels like we've been over this before, removing SMT is an efficiency loss, not a win. The catch is the loss will only exist in benchmarks, in most consumer workloads the number of threads available from the P+E combo will be high enough to out-scale the software. As for the area improvements, I would really wait and see for the benefits of removing SMT and AVX-512. There is a non-zero chance these features are just disabled in ARL.

While there is no doubt HT increases efficiency with a non-hybrid design, I'm not so sure about a hybrid design where we have cores specifically designed to handle ST (or low thread count) work loads and additional cores designed to handle high thread count workloads. We have a new paradigm with more variables to consider.

 

[coercitiv]

While there is no doubt HT increases efficiency with a non-hybrid design, I'm not so sure about a hybrid design where we have cores specifically designed to handle ST (or low thread count) work loads and additional cores designed to handle high thread count workloads. We have a new paradigm with more variables to consider.

Adding another set of cores in the mix does not change the outcome, the work done by the P cores can still be optimized. Consider you run a highly parallelized workload on a 8P+16E / 24T CPU, so SMT is disabled. The task is split between the P and E cores based on predetermined ratios, and the work done by the 8P/8T cores is finite. That same work can be done more efficiently by 8P/16T with lower clocks and lower voltage, as long as there's good scaling from 8T to 16T. The work done by the E cores is already accounted for.

I think SMT is the victim of a misunderstanding based on the power race in the recent years. Yes, SMT will increase power and temps when allowed to push the package power higher. However, when enforcing a sane power limit, SMT increases efficiency instead.

 

[igor_kavinski]

That same work can be done more efficiently by 8P/16T with lower clocks and lower voltage, as long as there's good scaling from 8T to 16T.

That certainly deserves some experimentation. Hulk is willing?

 

[coercitiv]

That certainly deserves some experimentation. Hulk is willing?

This was already done by someone in the forum. It's not the first time we're discussing this topic.