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

[Tigerick]

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 (20A)	Arrow Lake (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	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.)

 

[Wolverine2349]

I’m just realizing my 10700K is slower than those tiny e-cores

Oh way way way slower than Skymont I'm sure. But Gracemont its on par in non latency sensitive workloads and maybe a little ahead in INT, but your 10700K would spank the Gracemonts in latency and FP sensitive workloads.

Skymont e-cores will probably slaughter your 10700K in everything though.

 

[511]

I’m just realizing my 10700K is slower than those tiny e-cores

I am pretty sure those skymont will have MT equivalent to 12 Zen4 cores in MT Non avx 512 task

 

[MS_AT]

Not sure about that. More than 8 cores is not a must for gaming. But neither is more than 6. Some games start to stutter with 6 but 6 will run any game. More than 8 cores can help with some games today. There is a difference between beneficial and a must.

I suppose part of my hope is maybe Big.Little will actually not have scheudling issues this time since Skymont is supposed to be so good this time unlike Gracemont. The gap between IPC of Skymont and Lion Cove is so much closer than Gracemont and Golden/Raptor Cove that maybe all scheduling quirks of Big.Little will be fixed? AM I right on that?

Wasn't part of reason Big.Little caused scheduling issues with some games is because some games got caught on a Gracemont core and Gracemont had such much lower IPC and latency compared to Holden/Raptor Cove and game performance tanked as such. But with Skymont they should handle games as well as a 4.7GHz Golden Cove so if one gets on a core no big deal? Or not really and Big.Little has other scheduling problems besides just that. Or will the better thread director fix that as well.

Lion Cove will be faster, especially in gaming as it simply has more cache available per core compared to Skymont core. Is it 6x the L2 cache per core? [3MB vs 0.5MB] not to mention L1 and L0 caches. I mean this is simply the conclusion you can get from seeing that 7800X3D outperforms faster clocked cores, because it has more cache. Now the problem with scheduling is that ideally you keep the workload on the core it started running so the caches are primed. Every time you migrate it to another core the private caches are invalidated so you waste cycles to prime them again. Now if for some reason you move from P-core to E-core then well you end up with less cache and worse latency to boot It's up to Windows and game developers to ensure this doesn't happen as Windows has APIs to control this behavior.

I am pretty sure those skymont will have MT equivalent to 12 Zen4 cores in MT Non avx 512 task

Nah, the IPC comparison given by Intel in Spec was 1T, so the Skymont core had whole L2 cluster to itself. Start adding workloads, the cache per core drops. Not to mention in FP Skymont gains look impressive only because Gracemont was pathethic. So while it might come close, I would expect Zen4 to win due to higher per core caches, higher frequency.

 

[ikjadoon]

CUDIMM Standard Set to Make Desktop Memory a Bit Smarter and a Lot More Robust

www.anandtech.com

View attachment 107413

Did something go very wrong somewhere? So Arrow Lake IMC can't support DDR5-6400 at 1.1V and it will instead be considered overclocking???

Intel always lists the two-slot JEDEC maximum on its CPU spec sheets; meanwhile, four-slot motherboards cite the lower JEDEC speeds (due to the SI stubs of unpopulated DIMM slots).

We need to see a mini-ITX or CAMM2 Z890 board's JEDEC speeds, as that is what Intel will advertise.

Thus, I might suspect the highest-JEDEC (e.g., allegedly 6400) for ARL-S is likely reserved for 2-slot boards (e.g., mini-ITX), just as it was for Alder Lake & Raptor Lake. And maybe CAMM2 boards, too.

	Alder Lake	Raptor Lake	Arrow Lake
2-slot: 1DPC 1R	4800	5600	? [6400?]
2-slot: 1DPC 2R	4800	5200	? [6000?]
4-slot: 1DPC 1R	4400	4800	5600
4-slot: 1DPC 2R	4400	4400	5600
4-slot: 2DPC 1R	4000	4000	4800
4-slot: 2DPC 2R	3600	3600	4400
CAMM2	N/A	N/A	??

Major OEMs--the folks running JEDEC--have historically preferred 2x DIMM slots for cost savings, but that choice also improves signal integrity here.

 

[Wolverine2349]

Lion Cove will be faster, especially in gaming as it simply has more cache available per core compared to Skymont core. Is it 6x the L2 cache per core? [3MB vs 0.5MB] not to mention L1 and L0 caches. I mean this is simply the conclusion you can get from seeing that 7800X3D outperforms faster clocked cores, because it has more cache. Now the problem with scheduling is that ideally you keep the workload on the core it started running so the caches are primed. Every time you migrate it to another core the private caches are invalidated so you waste cycles to prime them again. Now if for some reason you move from P-core to E-core then well you end up with less cache and worse latency to boot It's up to Windows and game developers to ensure this doesn't happen as Windows has APIs to control this behavior.


Nah, the IPC comparison given by Intel in Spec was 1T, so the Skymont core had whole L2 cluster to itself. Start adding workloads, the cache per core drops. Not to mention in FP Skymont gains look impressive only because Gracemont was pathethic. So while it might come close, I would expect Zen4 to win due to higher per core caches, higher frequency.

By saying its up to Windows and game devs to ensure such behavior does not happen. SO can Windows control it. I mean if you lock the main thread to a P cores should there be no issue at all with Big.Little. If so how come Starfield has issues just with e-cores being on on ALder and Raptor even with Process Lasso? Is it not that simple or some games just bad behaved. Can Windows accommodate then?

And will games with Windows versions that had trouble with Alder and Raptor Lake still have the same exact trouble with Arrow Lake Big.Little. Or will the fact that Skymont cores even if slower than Lion Cove for gaming being so much better than Gracemont help mitigate the issues?

 

[511]

Lion Cove will be faster, especially in gaming as it simply has more cache available per core compared to Skymont core. Is it 6x the L2 cache per core? [3MB vs 0.5MB] not to mention L1 and L0 caches. I mean this is simply the conclusion you can get from seeing that 7800X3D outperforms faster clocked cores, because it has more cache. Now the problem with scheduling is that ideally you keep the workload on the core it started running so the caches are primed. Every time you migrate it to another core the private caches are invalidated so you waste cycles to prime them again. Now if for some reason you move from P-core to E-core then well you end up with less cache and worse latency to boot It's up to Windows and game developers to ensure this doesn't happen as Windows has APIs to control this behavior.


Nah, the IPC comparison given by Intel in Spec was 1T, so the Skymont core had whole L2 cluster to itself. Start adding workloads, the cache per core drops. Not to mention in FP Skymont gains look impressive only because Gracemont was pathethic. So while it might come close, I would expect Zen4 to win due to higher per core caches, higher frequency.

I forgot about frequency 🥲 but i guess at ISO frequency it should match it

 

[511]

Posting this in advance

 

[dullard]

By saying its up to Windows and game devs to ensure such behavior does not happen. SO can Windows control it. I mean if you lock the main thread to a P cores should there be no issue at all with Big.Little. If so how come Starfield has issues just with e-cores being on on ALder and Raptor even with Process Lasso? Is it not that simple or some games just bad behaved. Can Windows accommodate then?

And will games with Windows versions that had trouble with Alder and Raptor Lake still have the same exact trouble with Arrow Lake Big.Little. Or will the fact that Skymont cores even if slower than Lion Cove for gaming being so much better than Gracemont help mitigate the issues?

Developers get the ultimate say. Unfortunately, many developers are overworked and they may leave it up to the Windows scheduler. The Windows scheduler attempts to juggle everything (game in front, Windows background features, other programs like virus scanning, etc.) The Windows scheduler unfortunately isn't perfect in every combination of software and hardware.

The game developers really need to provide the scheduler information. When a thread is created, they can give the scheduler anything that ranges from gentle hints to outright specifying what exact core to use.

• Gentle developer hint, SetIdealProcessor tells the scheduler what core is desired whenever possible but has no strict restrictions: https://learn.microsoft.com/en-us/w.../nf-processthreadsapi-setthreadidealprocessor

• Gentle hint, SetThreadPriority tells the scheduler what threads are important and what threads are less important, but lets the scheduler do the rest: https://learn.microsoft.com/en-us/w...adsapi/nf-processthreadsapi-setthreadpriority

• Medium control, SetThreadInformation, tells the scheduler how to prioritize memory for the thread and whether or not to power throttle (that is, don't go idle in the middle of a boss fight): https://learn.microsoft.com/en-us/w...api/nf-processthreadsapi-setthreadinformation

• High control, SetThreadCPUSets, tells the scheduler which groups of cores to use for the thread (for example, only use P cores), but lets the power management still do what it wants: https://learn.microsoft.com/en-us/w...nf-processthreadsapi-setthreadselectedcpusets

• Max developer control, SetThreadAffinityMask, tells the scheduler to use a specific core, no matter what (the 2nd P core only or the 5th E core only). https://learn.microsoft.com/en-us/windows/win32/api/winbase/nf-winbase-setthreadaffinitymask

As more and more games get developed, they will be more and more written with the P cores and E cores in mind. When you run old games that didn't include this information, then you are most likely to run into E core scheduling issues.

 

[alcoholbob]

Posting this in advance

So you're saying Lion Cove is going to lose to the 7800X3D in gaming?

 

[511]

So you're saying Lion Cove is going to lose to the 7800X3D in gaming?

I meant to say Arrow lake will be mostly carried by skymont performance due to LNCs lack of HT

 

[ondma]

So you're saying Lion Cove is going to lose to the 7800X3D in gaming?

I would think so. 7800x3D is a few percent ahead of flat out 14900k, Going to the tile architecture and with a <10% 1T performance increase, I cant see ARL being more than a few percent at best faster than 14900K, and thus at best equal to 7800x3D. In fact, worst case, with the new architecture and minimal ST performance gain, I could see ARL showing no gain or even a slight regression vs RL.

 

[alcoholbob]

I would think so. 7800x3D is a few percent ahead of flat out 14900k, Going to the tile architecture and with a <10% 1T performance increase, I cant see ARL being more than a few percent at best faster than 14900K, and thus at best equal to 7800x3D. In fact, worst case, with the new architecture and minimal ST performance gain, I could see ARL showing no gain or even a slight regression vs RL.

If that's the case it would be really sad, because even the 14900K is ~3% faster in gaming with HT off, which closes some of that 7% gap with stock 7800X3D (which incidentally also is a bit faster in gaming with SMT off). So without a direct apples to apples comparison in terms of settings, ARL could claim higher gaming IPC when part of it came for free to start with.

 

[Wolverine2349]

If that's the case it would be really sad, because even the 14900K is ~3% faster in gaming with HT off, which closes some of that 7% gap with stock 7800X3D (which incidentally also is a bit faster in gaming with SMT off). So without a direct apples to apples comparison in terms of settings, ARL could claim higher gaming IPC when part of it came for free to start with.

It would be real sad. Though if its on par or only barely a regression, may be worth switching just to get more cores as some games are slowly starting to use them a bit. And on AMD side you get the dual CCD beyond 8 cores which has much worse scheduling issues and latency hit than Intel Thread Director and e-cores (in this case SKymont which should be pretty good) which works real well.

 

[Elfear]

That may be true. Though even if they do not clock well, if I can push 5GHz, I would be very happy with Golden Cove real world performance 10-12 P core chip HT disabled

No dealing with Big.Little hybird scheduling quirks if I can turly use Skymont and clock to 5GHz as GLC/RPC equivelent real world 99% gaming/other workloads

Do you really think a theoretical Skymont only CPU (ARL with P cores disabled), clocked to 5Ghz, and with ~RWC@5Ghz performance would outperform all of the current gen CPUs in gaming? It would be faster than Zen5, Zen43D, incoming full ARL, etc.? All because it has access to >8cores on the same ring? Any benchmarks you've seen that would lead you to that conclusion?

 

[Josh128]

You guys are getting lost in the weeds now. Why would Skymont outperform Lion Cove or even Zen 4 vanilla for that matter? That makes no sense, especially for 1 CCX Zen parts or P core only Intel parts.

I will say though, if Intels first desktop MCM chip can outperform 7800X3D in gaming, even with ARL having a node advantage, that will be truly impressive as AMDs 4th gen MCM chip cant do it.

 

[Wolverine2349]

Do you really think a theoretical Skymont only CPU (ARL with P cores disabled), clocked to 5Ghz, and with ~RWC@5Ghz performance would outperform all of the current gen CPUs in gaming? It would be faster than Zen5, Zen43D, incoming full ARL, etc.? All because it has access to >8cores on the same ring? Any benchmarks you've seen that would lead you to that conclusion?

For thread heavy games yes. Plus no HT and full access to 10-12 strong cores on a single die. No cross CCD latency. No Big.Little scheudling quirks. Best set and forget it solution for all games past 10 years and future games as well on a single box without using Process Lasso nor WIN11.

Hardware Unboxed had a video and some games benefit marginally form Ryzne 9 parts. Though some of that may be negated and some situations bad by cross CCD latency severe hit.

Skymont has as good of IPC as Raptor Cove supposedly.

If true with SKymont IPC and it translates to real word performance, an all Skymont CPU would mop the floor with vanilla Zen 4 and Zen 5 in gaming. Zen 4 X3D probably wins close in most games for now (though tuned DDR5 overclocking could negate 7800X3D advantage over Skymont), but stuck at 8 cores and if you want a best set and forget it solution for thread heavy games in niche cases plus some better responsiveness and productivity work all on one die without any process Lasso or scheduling issues to worry about, the all Skymont would be better being almost equal or slightly behind Zen 4 X3D and then Zen 5 X3D which will be like Zen 4 X3D if Zen 4 to 5 vanilla is an indication.

 

[Wolverine2349]

You guys are getting lost in the weeds now. Why would Skymont outperform Lion Cove or even Zen 4 vanilla for that matter? That makes no sense, especially for 1 CCX Zen parts or P core only Intel parts.

I will say though, if Intels first desktop MCM chip can outperform 7800X3D in gaming, even with ARL having a node advantage, that will be truly impressive as AMDs 4th gen MCM chip cant do it.

If Skymont truly does have Raptor Cove IPC or 2% better and it translates in 99% of all real world use cases including gaming and can hit almost 5GHz, Skymont would mop the floor or at least beat Zen 4 vanilla in gaming. The key word though is if it translates real world which we do not know if it does or will.

Though if it does especially with tuned XMP DDR5, it would hammer Zen 4 vanilla in gaming Vanilla Zen 4 i gaming is way behind well tuned Raptor Lake and Zen 4 X3D and even moderately behind Alder Lake with well tuned DDR5.

Now as for LIon Cove your probably right.

 

[Elfear]

For thread heavy games yes. Plus no HT and full access to 10-12 strong cores on a single die. No cross CCD latency. No Big.Little scheudling quirks. Best set and forget it solution for all games past 10 years and future games as well on a single box without using Process Lasso nor WIN11.

Hardware Unboxed had a video and some games benefit marginally form Ryzne 9 parts. Though some of that may be negated and some situations bad by cross CCD latency severe hit.

Ryzen 9 wins in some benchmarks vs Ryzen 7 due to slightly better boosting frequencies not because of a higher core count (AFAIK).

Skymont has as good of IPC as Raptor Cove supposedly.

If true with SKymont IPC and it translates to real word performance, an all Skymont CPU would mop the floor with vanilla Zen 4 and Zen 5 in gaming. Zen 4 X3D probably wins close in most games for now (though tuned DDR5 overclocking could negate 7800X3D advantage over Skymont), but stuck at 8 cores and if you want a best set and forget it solution for thread heavy games in niche cases plus some better responsiveness and productivity work all on one die without any process Lasso or scheduling issues to worry about, the all Skymont would be better being almost equal or slightly behind Zen 4 X3D and then Zen 5 X3D which will be like Zen 4 X3D if Zen 4 to 5 vanilla is an indication.

Yes but Raptor Lake CPUs boosted to 6Ghz. Zen43D, Zen5, Raptor Lake, and supposedly ARL are within spitting distance of each other. I don't see any way for a 5Ghz equivalent Raptor Lake to be better at gaming in this scenario. Unless you can point to benchmarks that show >8 cores bringing a benefit in gaming (and not just due to higher clocks from a higher tier CPU) me thinks you're tilting at windmills but I'm open to being wrong.

 

[MS_AT]

For thread heavy games yes. Plus no HT and full access to 10-12 strong cores on a single die. No cross CCD latency. No Big.Little scheudling quirks. Best set and forget it solution for all games past 10 years and future games as well on a single box without using Process Lasso nor WIN11.

Hardware Unboxed had a video and some games benefit marginally form Ryzne 9 parts. Though some of that may be negated and some situations bad by cross CCD latency severe hit.

Skymont has as good of IPC as Raptor Cove supposedly.

If true with SKymont IPC and it translates to real word performance, an all Skymont CPU would mop the floor with vanilla Zen 4 and Zen 5 in gaming. Zen 4 X3D probably wins close in most games for now (though tuned DDR5 overclocking could negate 7800X3D advantage over Skymont), but stuck at 8 cores and if you want a best set and forget it solution for thread heavy games in niche cases plus some better responsiveness and productivity work all on one die without any process Lasso or scheduling issues to worry about, the all Skymont would be better being almost equal or slightly behind Zen 4 X3D and then Zen 5 X3D which will be like Zen 4 X3D if Zen 4 to 5 vanilla is an indication.

I understand enthusiasm but at this pace, Skymont will turn out to be better than yet to be announced Apple M5. So if we start to pay attention to what intel claimed

Lunar Lake E-cores (codenamed "Skymont") IPC is on-par with Raptor Cove IPC on LLC/Ring.

under following conditions

Results are based on Intel's internal projections/estimates as of 5.13.2024(+/- 10% Margin of Error) on SPEC CPU 2017 rate baseline metrics (estimated), GCC12.1-O2 Linux (ISO) at Fixed Frequency (ISO) with similar memory + fabric for general (not vector heavy) workloads.

So first 2% IPC parity has +/- 10% error.

Next Intel used for comparison GCC12.1 at O2 optimization level. No further flags were given. Is therefore Intel trying to hide something? Nope I believe they are as honest as one can get, since Intel used GCC12.1 without specifying -march/mtune options the code will be tuned for baseline x64 target that iirc doesn't use anything more fancy than old SSE 2 or 4 at best. Read 128b per execution unit, so exceeding what Raptor Cove can do at that bit width, what places Raptor Cove at slight disadvantage as there are SPEC subtests that can be autovectorized by the compiler and if allowed to would make use of 256b SIMD units of Raptor Cove. But this is more relevant for non gaming workloads.

Last they are given the same fabric so the same amount of L3 cache. It's up for debate if they were therefore measuring 1 Skymont core from the cluster, so it had 2MB of L2 cache or not. I guess they did as these cores go in clusters so if it was meant to be representative it should be this. In reality especially for MT workloads each Skymont core on average will have 1/4 of this, so 0.5MB. This is more relevant for gaming workloads.

[source for intel quotes https://edc.intel.com/content/www/us/en/products/performance/benchmarks/computex-2024/]

So as you can see the IPC parity has very strict conditions.

We of course need to wait for real tests, but overhyping Skymont might end up in the same way as overhyping Zen5. Approach with caution and pat them on the back if they really deliver.

 

[jdubs03]

You mean the M4 right?

 

[Josh128]

If Skymont truly does have Raptor Cove IPC or 2% better and it translates in 99% of all real world use cases including gaming and can hit almost 5GHz, Skymont would mop the floor or at least beat Zen 4 vanilla in gaming. The key word though is if it translates real world which we do not know if it does or will.

Though if it does especially with tuned XMP DDR5, it would hammer Zen 4 vanilla in gaming Vanilla Zen 4 i gaming is way behind well tuned Raptor Lake and Zen 4 X3D and even moderately behind Alder Lake with well tuned DDR5.

Now as for LIon Cove your probably right.

Even if it would, did you forget that single CCX Zen 4 vanilla 7700X all cores in games to like 5.450 GHz? A 5.0 GHz Raptor Lake IPC chip is not going to beat that. It might tie it at best, but thats about it.

 

[cebri1]

Skymont won’t beat zen4 and RPL in normal conditions. But even if it is 10% slower at a fraction of the power, that still ADL performance which is still a very viable platform.

 

[cebri1]

BOE and Intel's Winning Display reduces power consumption by 65% using AI-driven 1 to 120 Hz dynamic refresh

Power-saving display tech is coming to Lunar Lake laptops.

www.tomshardware.com

Interesting, would be nice to have some outlets test this display.

 

[Wolverine2349]

Skymont won’t beat zen4 and RPL in normal conditions. But even if it is 10% slower at a fraction of the power, that still ADL performance which is still a very viable platform.

Hopefully and could be. Assuming one can even disable all P cores to begin with.

 

[Wolverine2349]

Even if it would, did you forget that single CCX Zen 4 vanilla 7700X all cores in games to like 5.450 GHz? A 5.0 GHz Raptor Lake IPC chip is not going to beat that. It might tie it at best, but thats about it.

I'd be pretty happy with that. Alder Lake IPC which is still very viable IPC and real world performance wise for gaming. Way better than Zen 3 and beats Zen 4 clock normalized and ties it or very close at a 300 to 400MHz clock speed deficit.