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.)

poke01 · Oct 26, 2024

adroc_thurston said:
DC variant of LNC at Al will still feature SMT.

Did Intel think Skymont would enough on client, so they got rid of it?

gdansk · Oct 26, 2024

poke01 said:
Did Intel think Skymont would enough on client, so they got rid of it?

In my opinion it is enough for client. 285K, 9950X, 14900K all have more MT than 99.9% of people need.
But SMT has so many advantages and Lion Cove for server needs AVX 512 and AMX and other changes that they may as well add SMT too.

H433x0n · Oct 26, 2024

gdansk said:
The curves we have for Cinebench R23 show it is efficient only at the TDPs they didn't ship it at.

Okay, but that wasn't the metric we've used the past few years, why move the goalposts?

Are we really saying that the only efficiency metrics that matter now are at 200W+?

gdansk · Oct 26, 2024

H433x0n said:
Okay, but that wasn't the metric we've used the past few years, why move the goalposts?

Are we really saying that the only efficiency metrics that matter now are at 200W+?

I didn't move the goalposts - Intel did in 2021 and AMD followed.
I don't know why Intel still feels compelled to ship only 160-250W CPUs with Arrow Lake where they are outside of their efficiency range afforded by lower platform power.

And at all power levels Lion Cove is less efficient than Zen 5 in SPECint despite being on a slightly more efficient manufacturing process. That's bleak. It didn't even achieve the main goal we were hoping for months ago after Zen 5% was revealed.

jpiniero · Oct 26, 2024

gdansk said:
I didn't move the goalposts - Intel did in 2021 and AMD followed.
I don't know why Intel still feels compelled to ship only 160-250W CPUs with Arrow Lake where they are outside of their efficiency range afforded by lower platform power.

Because these are the K parts. There'll be "65" and "35" W models announced at CES.

gdansk · Oct 26, 2024

jpiniero said:
Because these are the K parts. There'll be "65" and "35" W models announced at CES.

Do we know the PL2? Last gen "65W" was still in that range, 220W.

adroc_thurston · Oct 26, 2024

poke01 said:
Did Intel think Skymont would enough on client, so they got rid of it?

Pretty much yeah.

gdansk said:
And at all power levels Lion Cove is less efficient than Zen 5 in SPECint despite being on a slightly more efficient manufacturing process. That's bleak. It didn't even achieve the main goal we were hoping for months ago after Zen 5% was revealed.

yeah, what a waste of N3b wafers.

DavidC1 · Oct 26, 2024

Magio said:
So if Darkmont is a rather minimal upgrade as has been said, it's fair to expect Arctic Wolf and Golden Eagle (is that the first time this is mentioned?) to be big ones if they have to pave the way for the E core to take over the P core's duties after them.

Golden Eagle would also be the small upgrade.

poke01 · Oct 26, 2024

The question is will unified core be enough to take on Zen7, M7 and CPUs from ARM?
With currant pace of E-core I would say yes.

alcoholbob · Oct 26, 2024

H433x0n said:
Do we actually know it’s less efficient? I haven’t seen any mainstream testing data except for nT performance at PL2.

The only graphs I’ve seen of a perf/watt curve shows ARL doing better in power limited scenarios in cinebench R23 & R24.

The main thing I noticed is 285K is way more efficient than the dual CCD AM5 parts in gaming. When 7950X or 9950X are gaming, the power consumption is around 110-120W range, double that of the single CCD parts. But the 285K can run games at lower TDP (usually 50-70W) while still being in a dead heat with 9950X in productivity.

DavidC1 · Oct 26, 2024

gdansk said:
I'm not so sure about that. If my measurements are correct Zen 5C on N3E is about the same size as Skymont on N3B but it has SMT too which helps quite a lot in certain workloads. So it seems to waste less die space than letting Intel design a CPU core.

That's for Turin Dense right?

I'm getting 1.9mm2 without the L2 cache using SP5 package size which is 75.4mm x 72mm. That's bigger than Skymont with 1MB L2. Without it, it's 1.15mm2.

Loading…

www.amd.com

DrMrLordX · Oct 26, 2024

gdansk said:
But SMT has so many advantages and Lion Cove for server needs AVX 512 and AMX and other changes that they may as well add SMT too.

Intel isn't going to use Lion Cove in server, are they? Diamond Rapids is something else.

DavidC1 · Oct 26, 2024

DrMrLordX said:
Intel isn't going to use Lion Cove in server, are they? Diamond Rapids is something else.

10% gain over RWC is going to be awesome for a late 2026 CPU guys!

DrMrLordX · Oct 26, 2024

DavidC1 said:
10% gain over RWC is going to be awesome for a late 2026 CPU guys!

Yeah about that

...

*cough*

(also supposedly it uses Panther Cove, though it may already have been renamed)

adroc_thurston · Oct 26, 2024

DavidC1 said:
10% gain over RWC is going to be awesome for a late 2026 CPU guys!

but hey, 18A n stuffz.

biostud · Oct 26, 2024

alcoholbob said:
The main thing I noticed is 285K is way more efficient than the dual CCD AM5 parts in gaming. When 7950X or 9950X are gaming, the power consumption is around 110-120W range, double that of the single CCD parts. But the 285K can run games at lower TDP (usually 50-70W) while still being in a dead heat with 9950X in productivity.

Just remember the efficiency of the 3D vcache chips are usually way higher than the non vcache. Hopefully that will also be the case for Zen 5.

H433x0n · Oct 26, 2024

gdansk said:
I didn't move the goalposts - Intel did in 2021 and AMD followed.
I don't know why Intel still feels compelled to ship only 160-250W CPUs with Arrow Lake where they are outside of their efficiency range afforded by lower platform power.

And at all power levels Lion Cove is less efficient than Zen 5 in SPECint despite being on a slightly more efficient manufacturing process. That's bleak. It didn't even achieve the main goal we were hoping for months ago after Zen 5% was revealed.

Got it. Power scaling is irrelevant, the true way to measure efficiency is performance in the 200-250W range.

511 · Oct 26, 2024

H433x0n said:
Okay, but that wasn't the metric we've used the past few years, why move the goalposts?

Are we really saying that the only efficiency metrics that matter now are at 200W+?

The DLVR Screwing around at higher wattage that is clear from derbaeurs video

DavidC1 · Oct 26, 2024

511 said:
The DLVR Screwing around at higher wattage that is clear from derbaeurs video

DLVR is supposed to be energy efficiency at lower load and bursty workloads. It's a linear regulator so there will be losses at the high end. It's all described in the patent from years ago. You have two parallel regulators, so it can absorb the load better without much vdroop, so you can lower the average voltage.

511 · Oct 27, 2024

DavidC1 said:
10% gain over RWC is going to be awesome for a late 2026 CPU guys!

Diamond uses Panther cove which is an update over Cougar Cove which itself is update over Lion Cove

TSMC Nodes aren't easy to work with at High voltage

DavidC1 · Oct 27, 2024

511 said:
Diamond uses Panther cove which is an update over Cougar Cove which itself is update over Lion Cove

Cypress Cove: 18%
Willow Cove: 0%
Golden Cove: 19%
Redwood Cove: 0%(arguably minus)
Lion Cove: ~10%

Based on this trajectory Cougar Cove is the 0% shrink and Panther Cove is maybe 10%.

Do you know Exist is saying that 20A was so bad that it would have resulted in a worse Arrowlake?

gdansk · Oct 27, 2024

H433x0n said:
Got it. Power scaling is irrelevant, the true way to measure efficiency is performance in the 200-250W range.

It happens to be worse the way Intel ships it but I posted the SPECint graph a while back. It's worse across the entire range power range. And its performance per watt is worse across the Phoronix test suite too.

511 · Oct 27, 2024

DavidC1 said:
Cypress Cove: 18%
Willow Cove: 0%
Golden Cove: 19%
Redwood Cove: 0%(arguably minus)
Lion Cove: ~10%

Based on this trajectory Cougar Cove is the 0% shrink and Panther Cove is maybe 10%.

Do you know Exist is saying that 20A was so bad that it would have resulted in a worse Arrowlake?

The same exist50 said about 8% for IPC for Cougar Cove he hasn't given any projection for Panther Cove

As for 20A I don't know how bad it was except for the fact that it is a money looser and something went wrong with it that is why it was killed.Intel is not in a position for bleeding money on one off node without any significant returns which 20A failed to deliver

Maybe i will create a reddit account just to ask exist50 the IPC gains for PNC

Kocicak · Oct 27, 2024

gdansk said:
In my opinion it is enough for client. 285K, 9950X, 14900K all have more MT than 99.9% of people need.

Even a 14600K with HT turned off (14 threads) has more MT performance than 99,7% of people need.

It would be really unfortunatete if HT in the new CPUs was only turned off and no optimisations/improvements physical removal of this useless feature would allow were realised.

DAPUNISHER · Oct 27, 2024

AcrosTinus said:
Yes I am coping, I'm losing real money here

Your honesty is refreshing.

GN has their 265K review up. TLDW = "$404 value cannot be found."

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

Senior member

Attachments

Platinum Member

Diamond Member

Golden Member

Diamond Member

Lifer

Diamond Member

Diamond Member

Golden Member

Platinum Member

Diamond Member

Golden Member

Lifer

Golden Member

Lifer

Diamond Member

Lifer

Golden Member

Golden Member

Golden Member

Golden Member

Golden Member

Diamond Member

Golden Member

Golden Member

Super Moderator CPU Forum Mod and Elite Member