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]

Witha tdp of <15W i guess

 

[MarkPost]

Single core avx-512 will be the only win for 9950X vs 285K

Corona/vray uses intel embree which has avx-512 support so there is that as well blender doesn't support it and we are expecting 15-20% mt improvement so blender would be within 1-2%

AVX-512

AVX-512

forum.corona-renderer.com

Corona/VRay benchmarks gain nothing with AVX512. So Zen 5 perfomance with those rendering engines is not thanks to AVX512.

V-Ray (AVX512 ON in BIOS)


V-Ray (AVX412 OFF in BIOS)


Corona (AVX512 ON in BIOS)


Corona (AVX512 OFF in BIOS)

 

[511]

Corona/VRay benchmarks gain nothing with AVX512. So Zen 5 perfomance with those rendering engines is not thanks to AVX512.

V-Ray (AVX512 ON in BIOS)
View attachment 107071

V-Ray (AVX412 OFF in BIOS)
View attachment 107072

Corona (AVX512 ON in BIOS)
View attachment 107073

Corona (AVX512 OFF in BIOS)
View attachment 107074

Thanks can you run with a thread affinity to only 1 core ? and maybe we can ask a user in sub with 13900k to bench it maybe hulk

 

[MarkPost]

Thanks can you run with a thread affinity to only 1 core ? and maybe we can ask a user in sub with 13900k to bench it maybe hulk

Yes, give me some time. I can run them with my 13900K if no other user do it.

 

[Ghostsonplanets]

If 20 TOPS NPU is true, that means it won't even support Copilot+ 1st gen (minimum 40 TOPS).

And by that time in 2026, Microsoft would have unveiled Copilot+ 2nd gen (minimum 80 TOPS?).

Does Microsoft not want to bring Copilot+ to the sub-$500 laptop segment?

According to Bionic, it won't have a 40 TOPs NPU unfortunately. Such NPU cost quite a lot of die area, so I can understand why it would be greatly reduced for cheaper designs.

 

[cebri1]

P-cores only get about 22.5K in R23 at stock (correct me if Im wrong). Take 11% from that and you get ~20K. E-cores add about 18K for a total of ~40K.

Higher than 40K

With e-cores disabled, at 5.8Ghz, a 13900K can get around 24K points. Of course, you need to take into account the lack of HT which reduces the scores by about 25-30%, multiply by the IPC increase, and adjust for clocks (ARL-S does 5.4Ghz all cores). And Arrow Lake should be close to 18.5K 8 p-cores only.

Looking at the e-cores, RPC at 5.8Ghz with HT Off, should produce 2150~ points per core. Considering Skymont on ARL should have a similar IPC to RPC, I only adjusted for clocks (4.6Ghz all-e-cores) and I get 1700~ points per core, times 16 e-cores I get 27.5K points.

18.5K + 27.5K = around 46K at stock settings unless I made a mistake.

edit: Jaykihn got 43K+ on a QS sample. So between 44K and 46K is my final guess.

 

[DavidC1]

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

@igor_kavinski

It's not just about space, and Arctic Wolf should be 4x3, not 2x6, since they said 3-wide is fairly small. So Arctic Wolf would simply add another cluster.

Here's a comment from RWT: "Note 1: The above is missing the main point of having _multiple_ N-wide decoders: the average basic block (=BB) size is approximately 5 x86 instructions (4 linear instructions + 1 BB-terminating branch instruction) and if we assume that 50% of those terminating branches are at run-time resolved as TAKEN branches then it implies that _all_ x86 CPUs in the near future will be _required_ to have _multiple_ N-wide decoders."

For branches it's actually more performant going this way than a pure 8-wide, at least for x86. And description about Skymont says it works for loops too. Above 3-wide it runs into diminishing returns too, so clustering affords ILP beyond pure width increase.

So the loss in performance is not uniform, because in many cases it might even do better.

Combined with significant space and complexity savings, it's almost a no brainer.

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

Yes, but if they are going to widen the front end that much you can expect them to beef it up elsewhere, otherwise it would be a waste.

E cores stayed 2-wide between 50% perf/clock improvement. 2-3 wide they stayed for 70% improvement. 6-wide they stayed for another 70%.

 

[Josh128]

Higher than 40K

With e-cores disabled, at 5.8Ghz, a 13900K can get around 24K points. Of course, you need to take into account the lack of HT which reduces the scores by about 25-30%, multiply by the IPC increase, and adjust for clocks (ARL-S does 5.4Ghz all cores). And Arrow Lake should be close to 18.5K 8 p-cores only.

Looking at the e-cores, RPC at 5.8Ghz with HT Off, should produce 2150~ points per core. Considering Skymont on ARL should have a similar IPC to RPC, I only adjusted for clocks (4.6Ghz all-e-cores) and I get 1700~ points per core, times 16 e-cores I get 27.5K points.

18.5K + 27.5K = around 46K at stock settings unless I made a mistake.

edit: Jaykihn got 43K+ on a QS sample. So between 44K and 46K is my final guess.

Those 8 cores OC'ed to 5.8 are not representative of stock performance with E cores enabled, if you let them boost normally, its ~22k or so, thats why I used that number.

 

[cebri1]

Those 8 cores OC'ed to 5.8 are not representative of stock performance with E cores enabled, if you let them boost normally, its ~22k or so, thats why I used that number.

I think it's even better, as clocks should be more stable. In any case, adjusted for clocks, IPC and HT, I don't think my estimation will be off by much.

 

[MarkPost]

Thanks can you run with a thread affinity to only 1 core ? and maybe we can ask a user in sub with 13900k to bench it maybe hulk

ok here it is. First, 13900K (Intel Extreme Profile) MT.

13900K @ Intel Extreme Profile (Corona, MT)


13900K @ Intel Exterme Profile (V-Ray, MT)


And, ST.

13900K @5.7ghz (Corona, ST)


9950X @5.7ghz (Corona, ST)


13900K @5.7ghz (V-Ray, ST)


9950X @5.7ghz (V-Ray, ST)


EDIT

I have run Blender benchmark too:

13900K @ Intel Exterme Profile (Blender, MT)


9950X @Stock (Blender, MT)


13900K @5.7ghz (Blender, ST)


9950X @5.7ghz (Blender, ST)

 

[alcoholbob]

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

I'd be surprised if they couldn't with enough voltage. Even 9950X users with PBO on have reported being able to OC to at least 6GHz.

 

[poke01]

lunar lake in Dell, 268V

https://twitter.com/x/status/1832948909544710460

Cinebench ST: up to 124
Cinebench MT: 479

 

[poke01]

Some has to be wrong with it, how is slower than M2!!

 

[jdubs03]

Yeah those are very disappointing. On both ST and MT. My assumption was it was not plugged in. But still that MT score is awful, seems like it should easily be higher, I was expecting around 700-800.

 

[DokiDoki]

A Dell XPS laptop thermal throttling - why is this news?

 

[poke01]

A Dell XPS laptop thermal throttling - why is this news?

Ugh, yeah it’s a dell.

 

[alcoholbob]

A Dell XPS laptop thermal throttling - why is this news?

I guess for the fanless laptop fetish crowd...

 

[jdubs03]

I get the throttling for MT. But for ST?
Time to throw it in the freezer!

 

[511]

Why are we getting disappointed for ST Lmao intel said U9 gets 130 i subtracted 2 points from it(Marketing Tax) and it is 2% faster than 268V so 126 for estimate it get's 124 still good my issue is MT it sucks unless it's on battery it's Thermal throttling rip

 

[511]

ok here it is. First, 13900K (Intel Extreme Profile) MT.

13900K @ Intel Extreme Profile (Corona, MT)
View attachment 107081

13900K @ Intel Exterme Profile (V-Ray, MT)
View attachment 107082

And, ST.

13900K @5.7ghz (Corona, ST)
View attachment 107086

9950X @5.7ghz (Corona, ST)
View attachment 107085

13900K @5.7ghz (V-Ray, ST)
View attachment 107087

9950X @5.7ghz (V-Ray, ST)
View attachment 107084

EDIT

I have run Blender benchmark too:

13900K @ Intel Exterme Profile (Blender, MT)
View attachment 107116

9950X @Stock (Blender, MT)
View attachment 107117

13900K @5.7ghz (Blender, ST)
View attachment 107118

9950X @5.7ghz (Blender, ST)
View attachment 107119

9950X is 10% faster in st cs 13900k if LNC gets 14-16% IPC i guess both will be even in Single Core thanks for running the benches

 

[511]

Yeah those are very disappointing. On both ST and MT. My assumption was it was not plugged in. But still that MT score is awful, seems like it should easily be higher, I was expecting around 700-800.

I think 700 is way High it should be around 550-600 imo

 

[jdubs03]

Why are we getting disappointed for ST Lmao intel said U9 gets 130 i subtracted 2 points from it(Marketing Tax) and it is 2% faster than 268V so 126 for estimate it get's 124 still good my issue is MT it sucks unless it's on battery it's Thermal throttling rip

Ehh their slides put the 288V at 130-135. The 268V based on that should be higher than 124, something like 127-132. 124 is even with the X1E80 when it is supposed to be at least 5% more.

I think 700 is way High it should be around 550-600 imo

Ehh with 4 LNC and 4 Skymont, I was thinking you could get more than that. That we know depends on all-core clocks but seems possible to get significantly over 600.

If one p-core can get let’s just say 130 at 5.1Ghz. And all 4 of those with slightly lower clocks (maybe 4.8Ghz). Plus those 4 Skymont at 3.3Ghz+? Seems like you can get those numbers to add up to be over 700.

 

[511]

Ehh their slides put the 288V at 130-135. The 268V based on that should be higher than 124, something like 127-132. 124 is even with the X1E80 when it is supposed to be at least 5% more.

Ultra 9 get 130 not 132 127 would be the score at 5.0ghz for 130 considering the delta between OEM settings 124 is ok

Ehh with 4 LNC and 4 Skymont, I was thinking you could get more than that. That we know depends on all-core clocks but seems possible to get significantly over 600.

If one p-core can get let’s just say 130 at 5.1Ghz. And all 4 of those with slightly lower clocks (maybe 4.8Ghz). Plus those 4 Skymont at 3.3Ghz+? Seems like you can get those numbers to add up to be over 700.

I think all core power budget would be lower in multicore than in single core

 

[jdubs03]

Ultra 9 get 130 not 132 127 would be the score at 5.0ghz for 130 considering the delta between OEM settings 124 is ok

I think all core power budget would be lower in multicore than in single core

Right but Intel’s own slide is saying they get 1.165x the ST score of the HX 370, which has a score ranging from 112-116. That comes out to 130-135. The scores also match somewhere within the 1.0526x the X80-X84: 124-128.
I don’t think it’ll be that high (135) but they’ve put it out there that it could be. So a 127 is plenty reasonable for the 268V.

Spotted this. Not sure what to think of it, but kinda jives with what I was pointing out (could just be throttling, but that still seems too low).

https://twitter.com/x/status/1832999226751467745

 

[cebri1]

As always happens with laptop chips, results are going to vary widely based on cooling, size, screen type, conditions of the test, battery type, etc.