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.)

adroc_thurston · Oct 24, 2024

DavidC1 said:
tightening memory and cache latencies

You're not gonna shave many cycles out of things not L3. Like 4c L1 is the norm.
They do need to make a better L3$, the current one is absolutely dogwater.

DavidC1 said:
which will allow reducing of pipeline stages

Opcache hit pipe length is pretty short as is.

DavidC1 · Oct 24, 2024

Games such as Cyberpunk 2077 is probably an example where the high latency and reduced ring clock is affecting it a lot. 5-7% gains from 10% ring clocks, and in Raptorlake, it's 4.6GHz or 21% higher.

adroc_thurston said:
Opcache hit pipe length is pretty short as is.

I'm totally ignoring the OPCache, and that they go with Arctic Wolf/Unified Core with E core philosophy as a replacement.

GTracing · Oct 24, 2024

DavidC1 said:
And this:
View attachment 110232

Which is what they should be doing. Cut the clocks to ~5GHz, which will allow reducing of pipeline stages, tightening memory and cache latencies, and end up with a better core that's fit for desktop, server, and mobile.

Why do you think they had to back down ring clocks drastically? Cause they couldn't clock it that high without turning into Raptorlake.

I was more referring to the outliers where Arrow Lake loses to Alder Lake. It seems like a bug. http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...akes-discussion-threads.2606448/post-41322543

Though I agree the ring clock and accompanying latency definitely seem like the primary issue for gaming performance as a whole.

DavidC1 · Oct 24, 2024

GTracing said:
Though I agree the ring clock and accompanying latency definitely seem like the primary issue.

Intel said the new Thread Director stays on the E core until higher demands are required. And as we know, automated approaches don't work in all scenarios.

Combination of:
-Slow memory and ring
-Conflicts from scheduling
-Running on the E core

According to one Chinese review, the E core effect is in average about 5%. But there will be scenarios where the impact will be greater as games are quite sensitive.

igor_kavinski · Oct 24, 2024

@Markfw , I think this time you need to test 285K, just like you tested Alder Lake once before. Just pair it with a dirt cheap mobo and DDR5-6400 and I think it will give you good bang for the buck since it is beating 9950X in many Phoronix tests so it's possible that it does fairly well in DC workloads.

DavidC1 · Oct 24, 2024

Computerbase.de tests show Zen 5 gets 33% in CB2024 with SMT and 14900K gets 30%.

adroc_thurston · Oct 24, 2024

DavidC1 said:
I'm totally ignoring the OPCache, and that they go with Arctic Wolf/Unified Core with E core philosophy as a replacement.

Point is, average case pipe length isn't high.
Also Unified Core is trillion years away.

gaav87 · Oct 24, 2024

Anyone have latency numbers with e-cores turned off ? Maybe its, a miracle and 45ns with 9600mt/s ? xD

MangoX · Oct 24, 2024

Hitman928 said:
Server spend has shifted a huge chunk to cloud where AMD has more than 50% share. AMD has been slower to gain share in Enterprise, largely thanks to Intel offering huge discounts to try and hold onto share, but that has cratered their own revenue and margins and AMD has finally started picking up momentum in Enterprise anyway.

This. AMD is utterly dominating the Cloud and server hosting market. If you need a VPS/VDS or a dedicated server it's all Ryzen9 for frequency or Epyc for compute. SMT allows hosts to cheaply sell more vCPUs per node, and that allows more VM nodes per machine. As cool as Granite Ridge is, it has nothing against Turin dense for the cloud.

SiliconFly · Oct 24, 2024

desrever said:
My point is, people here act like E-core should replace P-core but they can't even clock to 4.6ghz without completely tanking efficiency.

Even if they can significantly improve IPC for the e-core next gen, if they are still clocked at <5 ghz good luck replacing P-core as a design.

Actually, in the future, if Arctic Wolf cores get another 30% perf uplift (at the same clock), their performance @ 4.6 GHz will be pretty much on par with a Lion Cove core running @ 5.7 GHz. Well, almost (like 90% to 95% minimum I think). Then it can be considered a viable replacement.

gaav87 · Oct 24, 2024

It seems that 8pcores, 8 e-cores is 5% faster then 8p/16e.
Guess 8p/8e works like 8p+ht... Great work intel

Joe NYC · Oct 24, 2024

Hitman928 said:
Do you mean Gamer's Nexus? They tested APO on and off but there was no difference for the 285k.

I guess I combined Hardware Unboxed with Gamer's Nexus into Hardware Nexus.

But you are right, it was in the Gamer's Nexus that showed APO ON and OFF

DavidC1 · Oct 24, 2024

gaav87 said:
It seems that 8pcores, 8 e-cores is 5% faster then 8p/16e.
Guess 8p/8e works like 8p+ht... Great work intel

View attachment 110234

Wow, now I'm thinking the weird P+E+P+E die config may be affecting performance. Because before in games you allocate things to P cores and they are all in one spot.

gaav87 · Oct 24, 2024

DavidC1 said:
Wow, now I'm thinking the weird P+E+P+E die config may be affecting performance. Because before in games you allocate things to P cores and they are all in one spot.

then there is no point in 285k for gaming. 265k with 8p/8e(4disabled) + OC

DavidC1 · Oct 24, 2024

gaav87 said:
then there is no point in 285k for gaming. 265k with for 8p/8e(4disabled) + OC

That's not the point, and 265K clearly performs quite a bit worse.

It's not just a matter of disabling clusters and being done with it. You have to figure out which clusters degrade gaming performance and which doesn't. Then a theoretical "gaming oriented" 265K would need to have dies disabled to take that into account.

These are the kind of details a well executing team would take into account. On a complex project as a modern SoC, there will be hundreds if not thousands of minute choices that all make a difference. And Arrowlake screwed up.

DavidC1 · Oct 24, 2024

Why skill matters:

Itanium: 10-stage pipeline, 800MHz clock, 300mm2 die, L3 cache off-die, 0.18u
Itanium 2: 8-stage pipelline, 1GHz clock, 420mm2 die with massive 3MB on die L3 cache, 0.18u, 2x the performance overall

Bonnell: 16-stage pipeline, 1.8GHz initial clock, 32nm, HT brought 30%
Silvermont: 13-stage pipeline, 2.4GHz clocks, 22nm, new uarch for 50%, at the same core size ISO-process. Intel said they removed HT and replacing it with OoOE and new uarch needed no increase in core size. 2x the overall performance. Regarding clocks, they attribute the increase entirely on 22nm.

There are kinds of details that we don't get to see public which make a real difference, and where high level disclosures will never explain.

GTracing · Oct 24, 2024

desrever said:
My point is, people here act like E-core should replace P-core but they can't even clock to 4.6ghz without completely tanking efficiency.

Even if they can significantly improve IPC for the e-core next gen, if they are still clocked at <5 ghz good luck replacing P-core as a design.

I don't think anyone is saying that the e-core should replace the p-core as-is. But rather that they should scale up the e-core so it's basically a new p-core.

Zen5c in Strix Point also has a lower clock ceiling and worse energy efficiency than the full Zen5 cores. An e-core with a less dense physical design could improve both of those aspects. And they can add cache, double the SIMD pipeline, and otherwise beef up the core.

I'm personally not 100% sold on the idea of e-cores taking over, but it seems like the best option for Intel in the long term. Maybe they'll do a fully redesigned core instead. Who knows. The one thing I'm sure of is that the current p-core is not good.

LightningZ71 · Oct 24, 2024

cannedlake240 said:
Or why not just let themselves get acquired by Qualcomm and then take over everything in a massive plot twist? Isn't Intel very good at politics

Mcdonald Douglas did that with Boeing. Look at how that's turned out...

desrever · Oct 24, 2024

GTracing said:
I don't think anyone is saying that the e-core should replace the p-core as-is. But rather that they should scale up the e-core so it's basically a new p-core.

Zen5c in Strix Point also has a lower clock ceiling and worse energy efficiency than the full Zen5 cores. An e-core with a less dense physical design could improve both of those aspects. And they can add cache, double the SIMD pipeline, and otherwise beef up the core.

I'm personally not 100% sold on the idea of e-cores taking over, but it seems like the best option for Intel in the long term. Maybe they'll do a fully redesigned core instead. Who knows. The one thing I'm sure of is that the current p-core is not good.

I mean, from the current implementation, the e-cores are still pretty meh in all metrics except area efficiency. The idea that a new core built with beefed up e-cores will actually be good doesn't have any more merit than a slim down p-core being better than the e-cores at everything.

I think Intel would hit the same problem in both approach in both directions. Their designs are just not that good, p-core or e-core.

LightningZ71 · Oct 24, 2024

gaav87 said:
It seems that 8pcores, 8 e-cores is 5% faster then 8p/16e.
Guess 8p/8e works like 8p+ht... Great work intel

View attachment 110234

I wonder how those 8 e-cores are being distributed? Are they activating just two quad core clusters, or are they distributing them evenly across all 4 clusters? The unified L2 in the quad core E-cores clusters can result in cache/memory contention issues for the E-cores when they get heavily used. Limiting to one or two cores per cluster gives much more L2 per core and more of each E-cores adjacent L3 slice per thread.

coercitiv · Oct 24, 2024

gaav87 said:
then there is no point in 285k for gaming. 265k with 8p/8e(4disabled) + OC

You would need all the E core clusters to maximize L2 per E core. See the comment above from @LightningZ71 for why that is likely. Also the L3 is bigger on 285K.

There may even be games where 8P4E works better than 8P8E, with 1 E core active per cluster. Presumably this is one of the things that Intel APO does - profiling the best active core configuration for each game.

DavidC1 · Oct 24, 2024

coercitiv said:
There may even be games where 8P4E works better than 8P8E, with 1 E core active per cluster. Presumably this is one of the things that Intel APO does - profiling the best active core configuration for each game.

This doesn't explain why 8P is worse than 8P8E though. Arrowlake apparently makes things even more complicated.

Did changing the core layout from PPPP-EEEE to PEPEPE really make a difference in thermals or clocks?

Arrowlake is just shoddily executed. After Meteorlake killed all the Raptors, now they are left with just Arrows, primitive civilization.

H433x0n · Oct 24, 2024

My standards were pretty low and I’m still disappointed. It seems Microsoft managed to mess up both AMD and Intel’s latest product launches with W11 issues.

There’s not a lot of definitive data on power scaling, was hoping ComputerBase would do that but no such luck. Would have been nice to see performance at 95W, 125W, 175W and PL2 for a variety of multicore tasks. All of the multicore data is at PL2 and way outside of efficiency zone.

Some of the 1T data is good (Python & JavaScript) but web browser performance is awful.

Hopefully the application performance gets cleaned up with Microsoft updates and platform maturity over the next few months but even if it does it’ll have horrible sales from the irredeemably bad gaming performance.

DavidC1 · Oct 24, 2024

desrever said:
I mean, from the current implementation, the e-cores are still pretty meh in all metrics except area efficiency. The idea that a new core built with beefed up e-cores will actually be good doesn't have any more merit than a slim down p-core being better than the e-cores at everything.

Sure it does. Because the design teams are different.

Also Skymont is pretty efficient on Lunarlake.

gdansk · Oct 24, 2024

H433x0n said:
My standards were pretty low and I’m still disappointed

This summarizes x64 desktop in 2024.

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

Senior member

Attachments

Diamond Member

Golden Member

Member

Golden Member

Lifer

Golden Member

Diamond Member

Member

Senior member

Golden Member

Member

Platinum Member

Golden Member

Member

Golden Member

Golden Member

Member

Golden Member

Senior member

Golden Member

Diamond Member

Golden Member

Golden Member

Golden Member

Diamond Member