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.)

yuri69 · Nov 6, 2024

DrMrLordX said:
That seems a bit late to market if they expect to compete with Turin dense.

Isn't Zen 6 like H2 2026?

511 · Nov 6, 2024

DrMrLordX said:
That seems a bit late to market if they expect to compete with Turin dense.

It is Q325 iirc it was mentioned in one of the calls i may be wrong though

SiliconFly · Nov 6, 2024

DrMrLordX said:
That seems a bit late to market if they expect to compete with Turin dense.

CWF was always a late 2025 product.

DrMrLordX · Nov 6, 2024

SiliconFly said:
CWF was always a late 2025 product.

My point still stands. It's Sierra Forest all over again.

yuri69 said:
Isn't Zen 6 like H2 2026?

Yeah but I'm not even talking about that (per se).

Long story short, Intel has nothing to face off against the competitor's dense solution for over a year. Once it shows up (Q3 or Q4 2025), it gets maybe two or three quarters to hopefully outperform Zen5c before it has to deal with Zen6c. Are we seeing Intel's problem yet?

511 · Nov 6, 2024

DrMrLordX said:
My point still stands. It's Sierra Forest all over again.

Yeah but I'm not even talking about that (per se).

Long story short, Intel has nothing to face off against the competitor's dense solution for over a year. Once it shows up (Q3 or Q4 2025), it gets maybe two or three quarters to hopefully outperform Zen5c before it has to deal with Zen6c. Are we seeing Intel's problem yet?

If it beats Zen5C than AMD doesn't have anything until Zen6C it is just a cycle Intel needs to break it

Henry swagger · Nov 6, 2024

Lion cove is a bloated for such low ipc.. skymont embarrassed idc in core design.. Stephen's team are fantastic.. the new United core should be great

mikk · Nov 6, 2024

DavidC1 said:
@GTracing What the heck?

Geekerwan overclocked the following:
-NGU from 2.6GHz to 3.2GHz
-D2D from 2.1GHz to 3.5GHz
-Ring from 3.8GHz to 4.2GHz

The E core latency went down from 194.2ns to 113ns or a 72% difference. D2D OC is 67%, so a slow D2D may be responsible for the super high latency. I would have liked him to test perf/clock after the overclocks.

CapFrameX tested it, it's 105ns for the E-core and 97 ns for the P-core. Latency on E-core <10% higher. I think this is a more accurate and realistic test.

https://twitter.com/x/status/1850266628652490777

511 · Nov 6, 2024

It is going to be Arrow to the knee lake on Desktop anyway

GTracing · Nov 6, 2024

mikk said:
CapFrameX tested it, it's 105ns for the E-core and 97 ns for the P-core. Latency on E-core <10% higher. I think this is a more accurate and realistic test.

https://twitter.com/x/status/1850266628652490777

That makes a lot more sense. I expect that is the "true" latency.

However, if the reason why Geekerwan measured higher latency is that the memory controller is in low-power mode (as João Bortolace suggested), then it's still possible that some synthetic tests are giving lower results than they should. Some single core benchmark that only run on one e-core could be affected by the increased latency.

DavidC1 · Nov 6, 2024

GTracing said:
However, if the reason why Geekerwan measured higher latency is that the memory controller is in low-power mode (as João Bortolace suggested), then it's still possible that some synthetic tests are giving lower results than they should. Some single core benchmark that only run on one e-core could be affected by the increased latency.

Things like these are what teams that are executing well and don't have to rush things out get fixed before releasing it and calling it a "product".

Geekerwan basically concluded that you could try many ways to improve performance but nothing really solves it. Sounds like the P core team also made Arrowlake.

DavidC1 · Nov 6, 2024

DrMrLordX said:
Long story short, Intel has nothing to face off against the competitor's dense solution for over a year. Once it shows up (Q3 or Q4 2025), it gets maybe two or three quarters to hopefully outperform Zen5c before it has to deal with Zen6c. Are we seeing Intel's problem yet?

Only if Zen 6 isn't on a 22 month cycle it has been previously. Because that puts us September 2026. Then Rogue River Forest would be less than 6 months away.

Regardless when you consider AMD C parts and Intel Forest parts are mainly for Cloud and VM workloads Sierra Forest is already doing pretty well.

igor_kavinski · Nov 6, 2024

GTracing said:
However, if the reason why Geekerwan measured higher latency is that the memory controller is in low-power mode (as João Bortolace suggested)

Why would they allow the memory controller to go into low power mode on desktop? They should release a microcode update ASAP to let users control this behavior and keep the memory controller up and ready for action ALL the time if they so wish.

OneEng2 · Nov 6, 2024

DavidC1 said:
@SiliconFly I doubt core counts are going to be that high. Process gains are really slowing down, and they'll need to spend whatever they get on a new uarch. No room for more cores, plus you run into scaling issues.

While I don't know what core counts will be on Zen 6, the CCD will have double the cores that Zen 5 has. How many CCD's the decide to put on a package seems like it will be only limited by the power the socket can provide.... and the socket power is reportedly going up drastically for Zen 6 DC boards.

511 said:
M5 is N3P from the rumours there is no way for N2 products before H2 2026 and AMD doesn't lead exactly like Apple so Zen6C is not happening before Q426

I disagree. We will most surely see N2 products in H1 (early H1 is my guess). Not sure how you are getting such a dismal forecast for TSMC.

511 said:
I am talking about products in hands of consumer like us or the server guys do you think there is a chance that we will have new N2 product in our hand by May/June considering the fact that it is a New node family not iteration like N3E and N3P

Absolutely. Why not? (ie May/June 2026). In fact, TSMC will be ramping up N2P by that timeframe.

DrMrLordX said:
That seems a bit late to market if they expect to compete with Turin dense.

My thought exactly.

yuri69 said:
Isn't Zen 6 like H2 2026?

Possibly. My guess is about mid year.

511 said:
It is Q325 iirc it was mentioned in one of the calls i may be wrong though

I remember this as well. I just don't believe it. I think the 18A process will slip by at least another quarter (You heard it here first!).

DrMrLordX said:
Long story short, Intel has nothing to face off against the competitor's dense solution for over a year. Once it shows up (Q3 or Q4 2025), it gets maybe two or three quarters to hopefully outperform Zen5c before it has to deal with Zen6c. Are we seeing Intel's problem yet?

That is how I see it as well. The cadence is off for Intel.

More to the point, none of this brings Intel back to profitability. To do that, they need to be selling foundry space on 18A .... and do it in high enough volumes to make their balance sheet work again. I think this is more problematic than the DC server chip design.

511 · Nov 6, 2024

OneEng2 said:
I disagree. We will most surely see N2 products in H1 (early H1 is my guess). Not sure how you are getting such a dismal forecast for TSMC.

What product tell me though Apple has been the lead product maker on TSMCs leading edge node the earliest i would take is june

OneEng2 said:
Absolutely. Why not? (ie May/June 2026). In fact, TSMC will be ramping up N2P by that timeframe.

My thought exactly.

Possibly. My guess is about mid year.

I remember this as well. I just don't believe it. I think the 18A process will slip by at least another quarter (You heard it here first!).

That is how I see it as well. The cadence is off for Intel.

Of course than it would mean TSMC is lying about N2 at IEDM 🙂

https://twitter.com/x/status/1853612313863205138

moinmoin · Nov 6, 2024

GTracing said:
it's still possible that some synthetic tests are giving lower results than they should.

If that's the case we would need other explanations for why performance in games is low and random.

511 · Nov 6, 2024

Anyways RIP Intel in Gaming it has been never been more over 🤣

DrMrLordX · Nov 6, 2024

511 said:
If it beats Zen5C than AMD doesn't have anything until Zen6C it is just a cycle Intel needs to break it

Right but Intel's track record up to this point hasn't been good on breaking this cycle. They're reacting slowly to competition rather than planning to beat it.

In theory, dense cloud solutions should be Intel's best datacenter products, given the rate at which Atom cores have been improving. In practice Intel's dense cloud solutions have been slow to market.

DavidC1 said:
Only if Zen 6 isn't on a 22 month cycle it has been previously. Because that puts us September 2026. Then Rogue River Forest would be less than 6 months away.

Given the way Gelsinger is plotting a course for Intel's future, there may not even be a Rogue River Forest. Also see above, don't expect Intel to execute on dates perfectly.

DavidC1 said:
Regardless when you consider AMD C parts and Intel Forest parts are mainly for Cloud and VM workloads Sierra Forest is already doing pretty well.

Sierra Forest is decidedly not doing very well. It can't compete with Turin dense on performance or perf/watt. It had only a few months of market relevancy, and now it's dead. The only way Intel is going to move this hardware is through the same methods that Intel managed to move any significant volume of Sapphire Rapids: vendor lock-in and discounts.

Hitman928 · Nov 6, 2024

511 said:
What product tell me though Apple has been the lead product maker on TSMCs leading edge node the earliest i would take is june

Of course than it would mean TSMC is lying about N2 at IEDM 🙂

https://twitter.com/x/status/1853612313863205138

If they mean TSMC won't reach HVM until 2026, then that's a delay. If they mean that TSMC will ramp into 2026, that was expected and doesn't prevent the first products from being taped out at the end of 2025.

Edit: The tweet doesn't match the IEDM link. Here is the quote from the session info:

jerrytw · Nov 6, 2024

511 said:
TSMC N2 mass production is 2026

https://twitter.com/x/status/1853608843235725622

https://twitter.com/x/status/1853612313863205138
Also found this graph
View attachment 111050

The 2026 timeline is for N2P.

511 · Nov 6, 2024

Hitman928 said:
If they mean TSMC won't reach HVM until 2026, then that's a delay. If they mean that TSMC will ramp into 2026, that was expected and doesn't prevent the first products from being taped out at the end of 2025.

Product tapes out 1-1.5 year before mass production usually 😅 as for 2nd part now i am confused by this I wil see the results directly in 2026

Hitman928 · Nov 6, 2024

511 said:
Product tapes out 1-1.5 year before mass production usually 😅

That's only for Intel.

511 · Nov 6, 2024

DrMrLordX said:
Given the way Gelsinger is plotting a course for Intel's future, there may not even be a Rogue River Forest. Also see above, don't expect Intel to execute on dates perfectly.

Why the negativity for Rouge river forest it is coming Raichu and Bionic squash has confirmed it on Twitter and can we stop blaming everything on Gelsinger outside of the GPU and his rear view mirror comment which is aging like fine wine he thought his predecessor has planned everything out just to see how horrible the products his predecessor created

DrMrLordX said:
Sierra Forest is decidedly not doing very well. It can't compete with Turin dense on performance or perf/watt. It had only a few months of market relevancy, and now it's dead. The only way Intel is going to move this hardware is through the same methods that Intel managed to move any significant volume of Sapphire Rapids: vendor lock-in and discounts.

How can you say that it is not doing well otherwise Intels DC revenue would not be 9% up they are due to SRF/GNR there is no other way for it to go up

511 · Nov 6, 2024

Hitman928 said:
That's only for Intel.

When do companies like AMD/Nvidia tape out than any public info?

Hitman928 · Nov 6, 2024

511 said:
When do companies like AMD/Nvidia tape out than any public info?

Depends on the product and the node. Something like MI300 will be longer between "tape out" and products shipping because there is so much advanced packaging and testing that has to happen. Something like a Ryzen product will be faster. I'd say anywhere from 6 - 9 months is typical. It could happen faster but typically not with these more advanced nodes. Intel seems to have a different meaning behind tape out than the rest of the industry (there are actually multiple tape outs that happen before a product is sold, usually when we talk about taping out in a public way, it means mass production. Intel seems to announce tape out for QS or even earlier).

511 · Nov 6, 2024

Server the home published SpecInt Performance for SRF/GNR

Ampere AmpereOne A192-32X Review A 192 Arm Core Server CPU

We dive into the Ampere AmpereOne A192-32X and discuss what this Arm-based CPU offers, how it performs, and what it means for competition

www.servethehome.com

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

Senior member

Attachments

Senior member

Golden Member

Golden Member

Lifer

Golden Member

Senior member

Diamond Member

Golden Member

Member

Golden Member

Golden Member

Lifer

Senior member

Golden Member

Diamond Member

Golden Member

Lifer

Diamond Member

Junior Member

Golden Member

Diamond Member

Golden Member

Golden Member

Diamond Member

Golden Member