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.)

511 · Thursday at 4:46 AM

alcoholbob said:
The AIDA64 latency doesn’t really explain why Arrow Lake is so bad at gaming, because they barely get any gaming improvement going from 90ns to 60ns or even sub-60 after tuning. A 60ns Raptor Lake build is still a good 20%+ faster than a 60ns 285K build in CPU-bound gaming situations. It’s definitely a confluence of multiple bottlenecks going on.

Why do we always check the DRAM Latency only. We need to check the Ring/LLC Latency and Bandwidth as well.

DrMrLordX · Thursday at 4:55 AM

MS_AT said:
I think the point I must have missed is anyone claiming that additional cores hurt gaming performance. I thought the point was they don't help, not that they actively hurt

Are we talking in circles? Look at the first graph. The whole point of it was to show, more cores = less performance on Factorio in direct refutation of the idea that more cores might actually help. In fact it's fairly neutral as is shown in the second graph. Though there's more to it than that:

https://www.reddit.com/r/factorio/comments/1ajdflz/how_many_cpu_cores_does_this_game_support

It really depends on what you build and how your pipes/belts/etc. interact with one another, if at all.

MS_AT · Thursday at 5:17 AM

DrMrLordX said:
The whole point of it was to show, more cores = less performance on Factorio in direct refutation of the idea that more cores might actually help

I understood it as, "look 7950X3d is not doing better here than 7800X3d, so more cores don't seem to help". Then we had 9950X3 vs 9800x3d that also don't seem to show the advantage of more cores. You took it to mean "more cores == less performance" for whatever reason, which no one claimed here to best of my understanding.

The thing is unless somebody will do a proper thread scaling test, we won't know, how exactly Factorio scales with threads taking into account all various OSes mechanics in play and how the devs may interact with the OS. But this is not a place to discuss that, nor I am a person to discuss that with, as I have 0 interest in Factorio.

igor_kavinski · Thursday at 6:02 AM

alcoholbob said:
A 60ns Raptor Lake build is still a good 20%+ faster than a 60ns 285K build in CPU-bound gaming situations. It’s definitely a confluence of multiple bottlenecks going on.

Source? That would be shocking to me.

Hitman928 · Thursday at 8:36 AM

DrMrLordX said:
Are we talking in circles? Look at the first graph. The whole point of it was to show, more cores = less performance on Factorio in direct refutation of the idea that more cores might actually help. In fact it's fairly neutral as is shown in the second graph. Though there's more to it than that:

https://www.reddit.com/r/factorio/comments/1ajdflz/how_many_cpu_cores_does_this_game_support

It really depends on what you build and how your pipes/belts/etc. interact with one another, if at all.

As the person who posted the first graph, you are wrong about the intention. The point was to show the game couldn't effectively use 16+ cores as was claimed, otherwise it wouldn't have mattered that it was being put on the wrong CCD or not.

As far as the reddit thread, it basically reiterates what I and others have been saying the whole time. Technically there are parts of the graphics engine that you can spread across many threads, but that becomes inconsequential as those parts are basically always waiting on the other parts of the engine that can't be split across more than a handful of cores at best.

Specifically any portion of the main game logic we identify that theoretically could be run in parallel is such a tiny fraction of the total update time that doing so just isn't worth it.

inquiss · Thursday at 10:43 AM

Hitman928 said:
As the person who posted the first graph, you are wrong about the intention. The point was to show the game couldn't effectively use 16+ cores as was claimed, otherwise it wouldn't have mattered that it was being put on the wrong CCD or not.

As far as the reddit thread, it basically reiterates what I and others have been saying the whole time. Technically there are parts of the graphics engine that you can spread across many threads, but that becomes inconsequential as those parts are basically always waiting on the other parts of the engine that can't be split across more than a handful of cores at best.

Exactly, and it makes sense, whether the game can use more than 8 cores or not, to keep the main threads on the one CCD for the quickest access to local L3 (specifically AMD, of course)

LightningZ71 · Thursday at 12:17 PM

As said up thread, Factorio is not the greatest game to be testing this with in the first place. Factorio's performance is HIGHLY dependent on how each factory is built. While a better CPU can let you get away with certain sins deeper into the game, eventually, it all comes down to efficient CPU resource usage by properly using the various elements in the game. Oddly, though expectedly, different CPUs can be better or worse in CERTAIN situations because different inefficiencies affect different aspects of the whole computer. I remember seeing whole threads about this on other forums. However, in general, it is still strongly ST performance bound.

dttprofessor · Thursday at 12:23 PM

I think foveros(36UM@MTL & ARL) makes the higher latency(70ns to ARL-S)，the speed for D2D(3.2GHZ @arl-S) is too slow ,and avoid more heat.
Nova lake put the imc @ soc tile too!!! Intel should find the better connection for chip2chip，maybe hybrid bonding ?

511 · Thursday at 12:49 PM

I find it so funny that we have the answer to arrow lake problem right in front of us and we are arguing about max core count in games 🤣

dttprofessor · Thursday at 3:52 PM

PTL will upgrade foveros-s to 25um ?

dullard · Thursday at 4:02 PM

dttprofessor said:
PTL will upgrade foveros-s to 25um ?

I think Lunar Lake is already at 25 um bump pitch.

DrMrLordX · Friday at 5:02 AM

Hitman928 said:
As the person who posted the first graph, you are wrong about the intention. The point was to show the game couldn't effectively use 16+ cores as was claimed, otherwise it wouldn't have mattered that it was being put on the wrong CCD or not.

That graph was not the one to post if that were your intention. The 7800X3D trounced the 7950X3D absolutely. Whereas the Zen5 graph didn't show that effect.

Hitman928 said:
As far as the reddit thread, it basically reiterates what I and others have been saying the whole time. Technically there are parts of the graphics engine that you can spread across many threads, but that becomes inconsequential as those parts are basically always waiting on the other parts of the engine that can't be split across more than a handful of cores at best.

Again, it depends on the build. Lots of people observe no scaling past 2-3 cores, whereas the developer indicates that builds with tons of unlinked pipes or belts could scale to 32 cores irrespective of the graphics engine.

dttprofessor · Friday at 5:32 AM

dullard said:
我认为 Lunar Lake 已经达到了 25 um 的颠簸间距。

是的

511 · Friday at 5:39 AM

Is it me or the messages are in Chinese

eek2121 · Friday at 5:52 AM

DrMrLordX said:
Are we talking in circles? Look at the first graph. The whole point of it was to show, more cores = less performance on Factorio in direct refutation of the idea that more cores might actually help. In fact it's fairly neutral as is shown in the second graph. Though there's more to it than that:

https://www.reddit.com/r/factorio/comments/1ajdflz/how_many_cpu_cores_does_this_game_support

It really depends on what you build and how your pipes/belts/etc. interact with one another, if at all.

The funny thing is, my own words were taken completely out of context a couple pages ago. I only claimed games can use more than 8 cores/threads. I kind of gave up after that, because it seems people want to argue that more cores = bad.

Factorio has a setting for this btw, and it used to default to a low number. I am not sure if it still does. There is no guarantee that any of the review sites thought to check this. There is one site that was mentioned a while back that showed the 9950X3D as being the fastest chip of them all. I will link it if I find it.

511 · Friday at 6:05 AM

eek2121 said:
The funny thing is, my own words were taken completely out of context a couple pages ago. I only claimed games can use more than 8 cores/threads. I kind of gave up after that, because it seems people want to argue that more cores = bad.

Factorio has a setting for this btw, and it used to default to a low number. I am not sure if it still does. There is no guarantee that any of the review sites thought to check this. There is one site that was mentioned a while back that showed the 9950X3D as being the fastest chip of them all. I will link it if I find it.

i think i have seen something like that somewhere as well Factorio has a setting to increase the size of the Map and on enabling it VCace looses it advantages it all depends on the parameters and the build setting for factorio.

Hitman928 · Friday at 9:01 AM

DrMrLordX said:
That graph was not the one to post if that were your intention. The 7800X3D trounced the 7950X3D absolutely. Whereas the Zen5 graph didn't show that effect.

It was plenty sufficient for its purpose in context. If the claim was that the game can use 16+ cores, showing 16 cores not being faster than 8 cores with the same architecture is a simple refutation of that claim. The fact that the 16 core CPU was slower was immaterial to the discussion in context. Additionally, there were other CPUs in that graph you could look at as proof as well that didn't have the "wrong CCD" issue.

DrMrLordX said:
Again, it depends on the build. Lots of people observe no scaling past 2-3 cores, whereas the developer indicates that builds with tons of unlinked pipes or belts could scale to 32 cores irrespective of the graphics engine.

You'll have to take it up with the developer as they already mentioned that going highly threaded doesn't help and, in fact, they have a blog post that trying to make the independent parts of the engine parallel actually hurts performance:

But these tasks are almost independent as neither trains or belts use electricity and belts don't interact with trains. There would be some details that would have to be solved, like Lua API calls and inserter activation order, but other than that it would be pretty straightforward. So I tried to run these things in parallel and measured how fast it would be:

To my surprise, the parallel version didn't speed things up, it was actually even slower. First I thought that it is caused by the fact, that the threads steal each others cache or the task is too short to be parallelized etc., but then I realized, that we already do the "prepare logic" in parallel. The prepare logic gathers all the data (sprite draw orders) for rendering and it is doing in parallel up to 8 threads. The task is iterating through the game world, and it is also quite short. Yet still the parallelization works pretty well there. After a few days (and many cppcon videos about multithreading later), I got the answer to my question. The answer is the same as many other answers related to performance, it is caused by cache invalidation.

eek2121 said:
The funny thing is, my own words were taken completely out of context a couple pages ago. I only claimed games can use more than 8 cores/threads. I kind of gave up after that, because it seems people want to argue that more cores = bad.

No one made the argument that more cores = bad, only that the game can't effectively use more than 8 cores. No proof was provided that it can, proof was provided that it can't.

eek2121 said:
Factorio has a setting for this btw, and it used to default to a low number. I am not sure if it still does. There is no guarantee that any of the review sites thought to check this. There is one site that was mentioned a while back that showed the 9950X3D as being the fastest chip of them all. I will link it if I find it.

If you have any proof that it can, please share because every time this gets brought up in the forum, it's a bunch of claims but no evidence. Everything from benchmarks to developer comments that I've seen on Factorio says that it can't effectively use more than a handful of cores at most.

jpiniero · Friday at 10:23 AM

Hitman928 said:
No one made the argument that more cores = bad, only that the game can't effectively use more than 8 cores. No proof was provided that it can, proof was provided that it can't.

Well, on mobile... more corez can be bad... especially with Intel's heat monsters.

511 · Friday at 10:29 AM

jpiniero said:
Well, on mobile... more corez can be bad... especially with Intel's heat monsters.

you are considering RPL as a normal thing

jpiniero · Friday at 10:35 AM

511 said:
you are considering RPL as a normal thing

It's been an issue since Intel went from 4 to 6 cores (heh)

511 · Friday at 10:40 AM

jpiniero said:
It's been an issue since Intel went from 4 to 6 cores (heh)

Part of that is on Intel's Node as well they were never optimized for low power and leakage until Intel 3 also it's not that bad with Meteor Lake and forward

511 · Saturday at 8:13 AM

A LNL Laptop with 5070

adroc_thurston · Saturday at 12:53 PM

511 said:
A LNL Laptop with 5070

This is the most pointless thing since TGL-H35. Bravo

eek2121 · Saturday at 3:09 PM

511 said:
i think i have seen something like that somewhere as well Factorio has a setting to increase the size of the Map and on enabling it VCace looses it advantages it all depends on the parameters and the build setting for factorio.

You can change the size of the map, but TBH I've never looked into it. Factorio has a very large map. The size isn't listed in the game, though I'm sure one could figure that out by searching.

At any rate, the setting I was referring to was regarding using threads for rendering (sorry for blown out colors, Windows AutoHDR messes up the colors in screenshots):

There are other games that do this as well. Not as many games use threading for other things, such as AI, however, to restate more clearly: There was a time when games weren't multithreaded at all, and a time where they could only use 2-4 cores. When 12-24 cores is common place, games will take advantage. Consoles will largely drive this.

More importantly. Most adults don't use their PC for gaming. It is used for a variety of workloads, some single threaded, some not.

511 · Sunday at 11:34 PM

Finally a good shot at Panther Lake

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

Senior member

Attachments

Platinum Member

Lifer

Senior member

Lifer

Diamond Member

Senior member

Platinum Member

Member

Platinum Member

Member

Elite Member

Lifer

Member

Platinum Member

Diamond Member

Platinum Member

Diamond Member

Lifer

Platinum Member

Lifer

Platinum Member

Platinum Member

Diamond Member

Diamond Member

Platinum Member