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

Tigerick · Aug 22, 2022

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

LightningZ71 · Friday at 8:14 AM

DavidC1 said:
xtor density metric doesn't really matter. You can't even compare them ISO anyway. Plus it varies tremendously with cache sizes.

Only thing that matters is the performance delivered per mm2. In that case Intel still sucks for Lion Cove and Battlemage. But Skymont for example is pretty damn good.

Which would really be weird since X3D is about gaming. What "ST" gaming is there nowadays?

While most big studio modern games are multi-threaded these days, the vast majority of games still have a single thread that carries a considerable amount of weight with respect to game load and performance dependence. ST performance is still VERY important in high framerate gaming. In games that aren't as demanding in that category, it doesn't really matter after a certain point.

eek2121 · Friday at 8:44 AM

Could've used 52 cores this morning

poke01 · Friday at 8:56 AM

eek2121 said:
Could've used 52 cores this morning

View attachment 125448

Yeah and it could have doubled as a space heater too..

511 · Friday at 9:10 AM

poke01 said:
Yeah and it could have doubled as a space heater too..

Depends entirely on TDP if it's ~300W without DLVR it is fine on desktop.

511 · Friday at 11:45 AM

For people who don't follow him
This is damm good.
Looks like they are actually making good progress for Panther Cove
@DavidC1

Io Magnesso · Friday at 11:50 AM

Nova lake is Coyote Cove Isn't it! ?

511 · Friday at 11:57 AM

Io Magnesso said:
Nova lake is Coyote Cove Isn't it! ?

Oh yeah it is renamed to Coyote Cove to avoid confusion.

Josh128 · Friday at 12:05 PM

511 said:
View attachment 125460

For people who don't follow him
This is damm good.
Looks like they are actually making good progress for Panther Cove
@DavidC1

Good, we need some blue competition.

jpiniero · Friday at 12:12 PM

Per two cores? Is Intel remaking Bulldozer?

511 · Friday at 12:16 PM

jpiniero said:
Per two cores? Is Intel remaking Bulldozer?

LMAO sharing L2 cache doesn't make a CPU bulldozer.

MS_AT · Friday at 12:31 PM

511 said:
LMAO sharing L2 cache doesn't make a CPU bulldozer.

Nope, but complicates scheduling

511 · Friday at 1:01 PM

MS_AT said:
Nope, but complicates scheduling

IF Microsoft is working on it than Yes 🤣

MS_AT · Friday at 1:15 PM

511 said:
IF Microsoft is working on it than Yes 🤣

Well, after Alder Lake release Linux was in worse condition, and scheduling for Threadrippers is still messed up in there unless somebody patched the scheduler after David Huang wrote about it.

Not to mention even if you you are an app developer and want to guide OS about thread allocation you have another corner case to account for.

511 · Friday at 1:34 PM

MS_AT said:
Well, after Alder Lake release Linux was in worse condition, and scheduling for Threadrippers is still messed up in there unless somebody patched the scheduler after David Huang wrote about it.

WE got so many patches after Alder Lake now Linux scheduler is pretty good imo but can't say that about the windows one.

MS_AT said:
Not to mention even if you you are an app developer and want to guide OS about thread allocation you have another corner case to account for.

Yup

Fjodor2001 · Friday at 1:34 PM

eek2121 said:
If it ends up being a Core 9 or whatever Intel calls their stuff these days, it will probably be $750-$800

So in that case much cheaper than 32C TR. Especially if you also take total system cost into account.

ondma · Friday at 2:14 PM

eek2121 said:
Core counts will only go up on specific products, and only because it is dead simple for them to add a second chiplet.

x86 isn’t going anywhere.

When you count hyperthreading, they will be pretty close in terms of performance. Power limits and IPC will be the deciding factor.

They won’t. It isn’t even confirmed to be a Core X part. It could end up being an Xeon workstation part.

If it ends up being a Core 9 or whatever Intel calls their stuff these days, it will probably be $750-$800

Here is my estimate of MT performance. 24 zen cores is 24 real cores plus 24 hyperthreading cores (estimate HT adds 25%), so that makes 30 "effective" real cores. If in fact NL has 16 plus 32, (assume E core 60% of P core-- that is 19 "P core equivalents"). So that leads to 35 "effective" real cores. So if they can get the power and thermals under control, I think Intel should have an edge in MT. Of course, that depends on NL P core matching or very close to Zen 6, and on how well the E cores perform. It also would depend on workload, i.e. how effectively it uses HT and the instruction set of the Intel E cores. Personally, it doesn't really matter to me, as my use case is gaming. In the interest of competition, though, it would be nice for Intel to lead in something, as I dont see them taking back the gaming crown from AMD v-cache chips.

Thunder 57 · Friday at 2:25 PM

Fjodor2001 said:
So in that case much cheaper than 32C TR. Especially if you also take total system cost into account.

If you believe that Intel (or anyone) would sell a CPU where each core costs $15 or less than sure. Especially when they are pushing for margins now and just announced more layoffs.

511 · Friday at 2:29 PM

ondma said:
Here is my estimate of MT performance. 24 zen cores is 24 real cores plus 24 hyperthreading cores (estimate HT adds 25%), so that makes 30 "effective" real cores. If in fact NL has 16 plus 32, (assume E core 60% of P core-- that is 19 "P core equivalents"). So that leads to 35 "effective" real cores. So if they can get the power and thermals under control, I think Intel should have an edge in MT. Of course, that depends on NL P core matching or very close to Zen 6, and on how well the E cores perform. It also would depend on workload, i.e. how effectively it uses HT and the instruction set of the Intel E cores. Personally, it doesn't really matter to me, as my use case is gaming. In the interest of competition, though, it would be nice for Intel to lead in something, as I dont see them taking back the gaming crown from AMD v-cache chips.

you need to modify your calculation an E core is 70% of a P core in Peak ST but in PL Workload it's lot closer this gap is going to be like to be lot less with NVL.

Thunder 57 said:
If you believe that Intel (or anyone) would sell a CPU where each core costs $15 or less than sure. Especially when they are pushing for margins now and just announced more layoffs.

The margin is Gross Margin and for new products also gross margin only includes Cost of Good Sold not R&D and other stuff

ondma · Friday at 2:48 PM

511 said:
you need to modify your calculation an E core is 70% of a P core in Peak ST but in PL Workload it's lot closer this gap is going to be like to be lot less with NVL.

The margin is Gross Margin and for new products also gross margin only includes Cost of Good Sold not R&D and other stuff

Hopefully you are right that the E core is closer to the P core in total throughput, but I was using a conservative estimate. Don't forget, the E cores have both an IPC and a clock speed deficit compared to the P cores. For instance in the 285K, the clock speed deficit alone is 20%.

coercitiv · Friday at 3:49 PM

511 said:
but in PL Workload it's lot closer this gap is going to be like to be lot less with NVL.

You also need to take into account the pressure on the L2 cache, when running heavy MT loads the E core has to work with less effective L2. That's one of the perks of the P core, their own L2 cache with a beautiful private alley to ringbus valley.

DavidC1 · Friday at 5:12 PM

511 said:
View attachment 125460

For people who don't follow him
This is damm good.
Looks like they are actually making good progress for Panther Cove
@DavidC1

Arrowlake's LNC is 3MB and 17 cycles. Why is 18 cycles at 4MB impressive?

ondma said:
(assume E core 60% of P core-- that is 19 "P core equivalents").

That's not true anymore since Arrowlake generation. Golden Cove was 40% faster per clock and clocked 33% faster in Alderlake and 35% faster in Raptorlake.

In Arrowlake Lion Cove is just 10% faster per clock and 24% higher clocks. It's already at 73% of the performance.

I assume Novalake will close the gap even further. Hence why they went to 2:1 ratio vs 4:1 ratio. Expecting 30% gains again on Arctic Wolf with two P core iterations maybe doing 10% each, so the P core will be just 5% faster. That's over 75% performance.

@LightningZ71 Why would an X3D part be slower in ST gaming?

poke01 · Friday at 8:55 PM

DavidC1 said:
Arrowlake's LNC is 3MB and 17 cycles. Why is 18 cycles at 4MB impressive?

Because it’s not. Nothing is impressive from Intel till Unified Core. Everything you see on Intel Twitter is hyped beyond comprehension as they are still living in Intels heyday.

adroc_thurston · Friday at 9:18 PM

poke01 said:
Nothing is impressive from Intel till Unified Core.

Atoms are good wrt perf@mm^2.
But that's CPU, at least Intel has a roadmap there.
GPU anything, especially GPGPU is lol.

LightningZ71 · Friday at 10:09 PM

DavidC1 said:
@LightningZ71 Why would an X3D part be slower in ST gaming?

My comment was directed specifically at the Gen on Gen performance difference between vanilla non-x3d Zen5 vs. vanilla non-x3d Zen6. There are only three cases where I would expect an X3D parts to be slower in ST performance than it's predecessor or the non-x3d sibling:
1- notable peak clock speed deficit, largely gone with Zen5.
2-Thermal throttling due to heavy MT loads running concurrently or poor cooling leading to heat soak. The vanilla part should generate slightly less thermal load and should maintain slightly higher clocks.
3- a weird corner case that exposes the minor latency hit that the 3d cache causes.

My argument for Zen6 is that, if the rumors are true, the 12 core CCX will have 48MB L3 cache at a comparable latency to the 8 core 32MB L3 CCX in Zen5. The 50% larger L3 would theoretically be available for a pure ST scenario, helping any apps that are dependent on it. It should also be less affected by cache pollution as the cache is larger and has more room to tolerate it with. Add in the expected 10% pic improvement from the rumor slide and it should be able to best Arrow Lake too.

511 · Friday at 10:26 PM

coercitiv said:
You also need to take into account the pressure on the L2 cache, when running heavy MT loads the E core has to work with less effective L2. That's one of the perks of the P core, their own L2 cache with a beautiful private alley to ringbus valley.

Yeah but not anymore going forward the private alley is going away 2P people have to share 😂.

adroc_thurston said:
Atoms are good wrt perf@mm^2.
But that's CPU, at least Intel has a roadmap there.
GPU anything, especially GPGPU is lol.

If only they weren't a bunch of idiot in Intel DC GPU space not cancelling everything.

DavidC1 said:
Arrowlake's LNC is 3MB and 17 cycles. Why is 18 cycles at 4MB impressive?

Why is it not that's1MB increase for 1 Cycle Skymont is 19 Cycles 4MB L2.

poke01 said:
Because it’s not. Nothing is impressive from Intel till Unified Core. Everything you see on Intel Twitter is hyped beyond comprehension as they are still living in Intels heyday.

Their Heydey died with 10nm delays lol.

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

Senior member

Attachments

Platinum Member

Diamond Member

Diamond Member

Platinum Member

Platinum Member

Junior Member

Platinum Member

Senior member

Lifer

Platinum Member

Senior member

Platinum Member

Senior member

Platinum Member

Diamond Member

Diamond Member

Diamond Member

Platinum Member

Diamond Member

Diamond Member

Golden Member

Diamond Member

Diamond Member

Platinum Member

Platinum Member