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

[Tigerick]

Wildcat Lake (WCL) Preliminary Specs

Intel Wildcat Lake (WCL) is upcoming mobile SoC replacing ADL-N. WCL consists of 2 tiles: compute tile and PCD tile. It is true single die consists of CPU, GPU and NPU that is fabbed by 18-A process. Last time I checked, PCD tile is fabbed by TSMC N6 process. They are connected through UCIe, not D2D; a first from Intel. Expecting launching in Q2/Computex 2026. In case people don't remember AlderLake-N, I have created a table below to compare the detail specs of ADL-N and WCL. Just for fun, I am throwing LNL and upcoming Mediatek D9500 SoC.

		Intel Alder Lake - N	Intel Wildcat Lake	Intel Lunar Lake		Mediatek D9500
Launch Date		Q1-2023	Q2-2026 ?	Q3-2024		Q3-2025
Model		Intel N300	?	Core Ultra 7 268V		Dimensity 9500 5G
Dies		2	2	2		1
Node		Intel 7 + ?	Intel 18-A + TSMC N6	TSMC N3B + N6		TSMC N3P
CPU		8 E-cores	2 P-core + 4 LP E-cores	4 P-core + 4 LP E-cores		C1 1+3+4
Threads		8	6	8		8
Max Clock		3.8 GHz	?	5 GHz
L3 Cache		6 MB	?	12 MB
TDP		7 W	Fanless ?	17 W		Fanless
Memory		64-bit LPDDR5-4800	64-bit LPDDR5-6800 ?	128-bit LPDDR5X-8533		64-bit LPDDR5X-10667
Size		16 GB	?	32 GB		24 GB ?
Bandwidth			~ 55 GB/s	136 GB/s		85.6 GB/s
GPU		UHD Graphics		Arc 140V		G1 Ultra
EU / Xe		32 EU	2 Xe	8 Xe		12
Max Clock		1.25 GHz		2 GHz
NPU		NA	18 TOPS	48 TOPS		100 TOPS ?

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.

 

[LightningZ71]

I still don't fully grasp why we're having a heart attack over HT on arrow lake? It would add, effectively, the throughput of maybe two additional E or P cores. Are these processors really so crippled in MT performance without HT on the P cores that it's a massive issue? The scores in MT are likely being held back FAR more by the L3 latency numbers than anything else, and the added contention from a second thread running on the P cores would probably make that situation worse.

 

[Hulk]

I still don't fully grasp why we're having a heart attack over HT on arrow lake? It would add, effectively, the throughput of maybe two additional E or P cores. Are these processors really so crippled in MT performance without HT on the P cores that it's a massive issue? The scores in MT are likely being held back FAR more by the L3 latency numbers than anything else, and the added contention from a second thread running on the P cores would probably make that situation worse.

I was literally going to type, "Why are we discussing/arguing MT performance on ARL when there is something else going on that is much more important."

As you correctly note L3 latency seems to be at the heart of the problem with ARL.

The reviewers are "assembly line" pushing out the reviews with the same old kind of useless data.

What would be interesting?
Isolate the Skymont cores and really test them.
Same for Lion Cove and try to discover why certain apps that performed well in Raptor Cove are not doing well in Lion Cove.
Using the same air cooler for Raptor and Arrow test at the same power/core temp. What do those real world numbers look like? Repeat for a 240 and 360 AIO.

Most of the performance of a 14900K can't be obtained without serious cooling, tuning, and a good sample. If ARL is more plug-and-play then there is value there that is yet not advertised.

 

[GTracing]

I still don't fully grasp why we're having a heart attack over HT on arrow lake? It would add, effectively, the throughput of maybe two additional E or P cores. Are these processors really so crippled in MT performance without HT on the P cores that it's a massive issue? The scores in MT are likely being held back FAR more by the L3 latency numbers than anything else, and the added contention from a second thread running on the P cores would probably make that situation worse.

Yeah, all the belly-aching about hyper threading is ridiculous. There are several bigger issues with Arrow Lake.

 

[Kocicak]

But, when someone buys a desktop computer, that alone makes it not average. Everyone I know personally who has a desktop is using it for gaming.

Have you tested how few cores are really needed for gaming? I did that recently, I was cutting cores and lowering frequency, and my 4070 requires ONLY 6 pretty slow cores (without HT).

 

[MoistOintment]

How come it took until Arrow Lake to double the iGPU's performance? Why couldn't he release a better iGPU with 14th gen?

Why didn't they redesign and tape out a new design for RPL refresh that included porting Alchemist to Intel 7? Why in the world would they do that?

14th gen was just what they had to work with after it became apparent that MTL's issues would make for an even worse desktop launch than ARL's. Porting Alchemist to Intel 7 and redesigning RPL dies to accommodate it wouldn't have even been brought up in the discussions.

 

[igor_kavinski]

I don't understand how HT would have made ARL better?

Intel Core Ultra 200S Content Creation Review

We examine the performance of Intel's new Core Ultra 200S series desktop processors in content creation applications.

www.pugetsystems.com

Check out how well 285K is doing in some Adobe tests and all Unreal tests save one. HT could've tipped almost all of these benches in ARL's favor. It would've definitely become slightly more power hungry but Intel suddenly pretending that they care about power efficiency is like an obsessive gambler throwing away the night's considerable winnings just because he got struck by a sense of morality. IF the HT switch is there somewhere inside the microcode firmware, I expect Intel to switch it on once they see how badly ARL sales are going, to try to at least tempt the productivity users, if they haven't already chosen Zen 5.

For example, from my CB R23 MT data for HT/SMT below while AMD has had better implimentations, generation-to-generation Intel has only been a few percentage points behind.

You need one more real world MT workload in your arsenal to figure out the MT efficiency of CPUs. Relying on just CB R23 can lead you to possibly incorrect conclusions about the general MT performance of different CPU architectures.

 

[igor_kavinski]

Have you tested how few cores are really needed for gaming? I did that recently, I was cutting cores and lowering frequency, and my 4070 requires ONLY 6 pretty slow cores (without HT).

You can't claim that without showing us data for over 20 games at the minimum, from different genres. And you are making your claim with Nvidia drivers that are notoriously CPU limited.

 

[igor_kavinski]

the added contention from a second thread running on the P cores would probably make that situation worse.

Let the user decide what he wants to do. If Intel did indeed switch off HT silicon present in the Lion Cove cores, they should provide a BIOS option to turn it on with a disclaimer that they are not responsible for any errors or financial loss resulting from using HT.

 

[Tup3x]

Handbrake / video encoding, audio encoding, OBS, compiling are some great examples of MT workloads.

Audio encoding is a fine example of workload that is single threaded. I don't know a single multi threaded audio encoder other than some proof of concept level stuff. Or what did you mean exactly? Multiple streams/files at once?

OBS... well, that depends.

 

[igor_kavinski]

If Intel did indeed switch off HT silicon present in the Lion Cove cores

By the way, someone really dumb is there inside Intel who keeps wasting the hard work of engineers by switching stuff off at the last minute. AVX-512 was the same story. We just don't have proof that HT was available at the ES or QS stage.

 

[StinkyPinky]

This may be the worst launch Intel has done since the Pentium 4. Granted Haswell to Kaby Lake weren't up to much, but at least it didn't freaking regress.

 

[Jan Olšan]

Ian Cutress said it the best in his tweets, Intel was never in position to satisfy the tech reviewer crowd.

What that crowed wanted:

• A CPU from Intel that is faster than the AMD 9950X
• Faster than the 7800x3D and potentially significantly faster so that the upcoming 9800x3D is within rage as well.
• While consuming as much energy as the 7800x3D, a chip with half its guts missing and just a fraction of the MT performance.

You see how constrained the whole tech media space is, in any case Intel would get negative press.

Scenarios:

• Intel goes for the performance again, with a higher base TDP, aggressive clocks and very high power draw. You get in all cases better performance than the 14900K
  • Result: It would be compared to the 7800x3D, we would get 100s of videos with a flames in the thumbnail. No one would even acknowledge the performance, all would focus on the power draw in order to negate the competitive performance while ignoring the MT performance. The fate of the 13900K and 14900K would repeat itself.
• Intel goes for a imaginary skew that uses the old Intel 7 and only 8 Lion Cove cores with HT enabled and your favorite fairy dust Level 4 Cache with 100MB. Let us say it just exceeds the 7800x3D by 5%.
  • Result: The power draw, now of course MT performance would matter to reviewers, they will start benchmarking shader compilation.
  • They will remind you in every step that the platform might never get a new CPU and that AMD will last until the heat death of the galaxy. (Intel always has 2 to 3 gens and never ever has confirmed in the past)
  • They will start discouraging people from considering it by mentioning the 9800x3D that has not launched and plastering frame per watt graphs everywhere. There might be an additional disclaimer to tell people that even though this Intel 7 Gen3 is said to be safe every reviewer would tell people to not trust it and that it is a old 10NM node in reality that has failed already.
• Intel goes for efficiency with no performance gain.
  • Result: mention that the platform is dead, say that AMD is better due to AMD lasting until the end of time. Compare it to the 7800x3D again, then mention the 9800x3D and tell everyone to not buy it.
  • DOOMSDAY:_
    • Go for efficiency with performance regression:
      • Everything mentioned above.
      • There was never a world where Intel would get positive media coverage but meeting the crazy standard mentioned at the start.
      • As they should because any imaginary world I mentioned is better than where we ended up.

Really? He tries to paint it as if it is just a problem of selecting core count and power targets and if the power was high enough or E-Cores removed (At cost of MT performance), the problems would disappear. That IS NOT TRUE.

Arrow Lake's problem is the interconnect or memory latency killing gaming performance - this would not be fixed.
Arrow Lake's problem is kind of small IPC gain on P-Core - this would not be fixed, more power would not improve ST boost frequencies, it would only push cinebench even higher.
AMD's cache advantage (games only) would not go away no matter what.

I mean, I try to defend these processors from the unrealistic bashing, but this is unrealistic cope/whitewashing. People may be bashing Intel to undeserved degree or for things that are not bad. But those "scenarios" is almost like something marketing would try to say to excuse the actual problems.

 

[Josh128]

Audio encoding is a fine example of workload that is single threaded. I don't know a single multi threaded audio encoder other than some proof of concept level stuff. Or what did you mean exactly? Multiple streams/files at once?

OBS... well, that depends.

FLAC encoding is multithreaded.

flac: Encodes, decodes and tests FLAC files – The Geek Diary

flac is a command-line tool that provides functionality for encoding, decoding, and testing FLAC (Free Lossless Audio Codec) files. FLAC is a popular audio format known for its lossless compression, which means it retains the original audio quality while significantly reducing the file size. The...

www.thegeekdiary.com

 

[LightningZ71]

Let the user decide what he wants to do. If Intel did indeed switch off HT silicon present in the Lion Cove cores, they should provide a BIOS option to turn it on with a disclaimer that they are not responsible for any errors or financial loss resulting from using HT.

My point is that HT being enabled on the Arrow Lake P cores in the chip's current condition may not even help throughput in any meaningful way on the vast majority of apps. The L3 latencies being what they are, both threads will likely spend all their time stalled for memory access, and the additional requests on it will slow everything else down with it. We also know that HT isn't free from a thermal perspective. How would the CCD even behave with the additional heat being dumped into it from even hotter P cores? Does it hit other limits?

 

[Hulk]

Intel Core Ultra 200S Content Creation Review

We examine the performance of Intel's new Core Ultra 200S series desktop processors in content creation applications.

www.pugetsystems.com

Check out how well 285K is doing in some Adobe tests and all Unreal tests save one. HT could've tipped almost all of these benches in ARL's favor. It would've definitely become slightly more power hungry but Intel suddenly pretending that they care about power efficiency is like an obsessive gambler throwing away the night's considerable winnings just because he got struck by a sense of morality. IF the HT switch is there somewhere inside the microcode firmware, I expect Intel to switch it on once they see how badly ARL sales are going, to try to at least tempt the productivity users, if they haven't already chosen Zen 5.


You need one more real world MT workload in your arsenal to figure out the MT efficiency of CPUs. Relying on just CB R23 can lead you to possibly incorrect conclusions about the general MT performance of different CPU architectures.

I disagree. The tests that the 285K loses look to be architecture related not thread starved.

Yeah, I had the data I presented readily available. More is always better, but some is better than none.

 

[dttprofessor]

Why intel reduce ring frequency on ARLs? For efficency?

 

[OneEng2]

I think Kocicak is being taken out of context here.
The point being made is that modern Intel CPU's with 12 or more cores don't really benefit from HT except maybe for a few niche cases.
This has certainly been my experience with my 14900K. The 16 physical E cores are much more useful than the 8 logical cores, which only serve to heat the core beyond what my cooler can handle.

Also ARL is proof. No HT and very good MT performance. Ironically it is more the applications that don't employ all of the cores are the ones that don't perform as well as we'd hoped.

From an architectural standpoint, adding SMT adds more performance per area than adding a new core. It is also more performance per watt compared to another core.

My point is that it is a good architectural element in processors designed for advanced operating systems and advanced programs. For gaming only, I agree with you. If that is the argument, then why have eCores at all? Why not just 4 very powerful P cores with a ton of cache and call it done?

Additionally, there is the idea of using a different core architecture for P and E cores. I am still on the fence with this one; however, I think it is very advantageous to have SMT on both regardless of if the P and E cores are different core designs or not simply because of the die and power efficiency SMT brings. This is especially important for the DC market ..... and unless we are talking about having P cores and E cores for desktop and another design for P cores and E cores in the server (for Intel that would be 4 different core design teams for a single chip), this doesn't seem like a great strategy for a company that is talking about laying off 50% of its engineers IIRC.

I feel like you guys have very different definitions of what an "average consumer" looks like which is causing some friction.

If someone forced my hand, I would say the height for an average consumer is paying for bills via online banking, maybe.
Everything has moved to phones.

But, when someone buys a desktop computer, that alone makes it not average. Everyone I know personally who has a desktop is using it for gaming.
My social circle is not representative though.

Very good point. Desktop is used for more heavy workloads today than in the past. People are likely to browse the internet on their phone. Pretty much anything that is simple to do, and doesn't require lots of screen space to work well is going to be done on a phone app.

Heavy workloads are very often heavily MT with the exception of games.

I would argue that gamers should have a specific core variant that just has some really fast P cores with lots of cache. No need for e Cores for gaming.

But all of that depends on how much cost HT or SMT adds? we know an E core is like 1/3 of a P core in space, how much is the area cost of HT/SMT? and the power efficiency and performance too of course. Only then it makes sense to compare. I've read it's about 5% of the P core area, though of course it may depend on which exact core.

On the surface, HT/SMT seems more efficient per area than E cores, that's in the case one has to choose between one or the other of course, you can do both which should already be the case with Lion Cove. But then HT also limits you very quickly, in the fact that if you have 8 cores, and add HT, then you cannot get more than the 25-30% extra you get, unlike E cores that you can keep adding more of them and gain a lot of performance.

Exactly.

I still don't fully grasp why we're having a heart attack over HT on arrow lake? It would add, effectively, the throughput of maybe two additional E or P cores. Are these processors really so crippled in MT performance without HT on the P cores that it's a massive issue? The scores in MT are likely being held back FAR more by the L3 latency numbers than anything else, and the added contention from a second thread running on the P cores would probably make that situation worse.

For gaming (and likely some other applications) you are correct; however, see my discussion above. The architecture of a new generation specifically targeted for the desktop and laptop? What about DC? What about work station?

Is it even possible for Intel to survive on desktop and laptop sales alone? I don't think so.

 

[igor_kavinski]

I disagree. The tests that the 285K loses look to be architecture related not thread starved.

Maybe it's not thread starvation that is limiting 285K in those tests. Maybe it's their 8 wide decode hardware sitting idle because there isn't an extra stream of instructions coming from the virtual thread.

https://course.ece.cmu.edu/~ece742/f12/lib/exe/fetch.php?media=elmoursy_frontendsmt03.pdf

Second, in an SMT processor, the inability of a single thread to utilize the issue queue (for instance, due to an Icache miss) can be made up by filling the queue with instructions from other threads. The result is increased utilization of the issue queue relative to a singlethreaded machine,

How would one go about accomplishing the above increased utilization of a core's front-end without an additional virtual thread? If there was a way, Intel certainly didn't seem to implement it in Arrow Lake.

 

[igor_kavinski]

Why intel reduce ring frequency on ARLs? For efficency?

They were most likely scared because of the Raptor Lake ring degradation fiasco.

 

[OneEng2]

They were most likely scared because of the Raptor Lake ring degradation fiasco.

Not sure, but some reports suggest they may have issues anyway:

Things look dire for Intel as Arrow Lake reportedly has instability issues

Stop me if you've heard this one before.

www.xda-developers.com

Arrow Lake instability issues?? RUH ROH!!

Like everyone else, been looking for some news of the new Arrow Lake chips and in my news feed without needing to search found an article about Intel having instability issues early in the game. https://www.xda-developers.com/arrow-lake-reportedly-has-instability-issues/ Not they haven't had...

www.techpowerup.com

 

[dttprofessor]

From an architectural standpoint, adding SMT adds more performance per area than adding a new core. It is also more performance per watt compared to another core.

My point is that it is a good architectural element in processors designed for advanced operating systems and advanced programs. For gaming only, I agree with you. If that is the argument, then why have eCores at all? Why not just 4 very powerful P cores with a ton of cache and call it done?

Additionally, there is the idea of using a different core architecture for P and E cores. I am still on the fence with this one; however, I think it is very advantageous to have SMT on both regardless of if the P and E cores are different core designs or not simply because of the die and power efficiency SMT brings. This is especially important for the DC market ..... and unless we are talking about having P cores and E cores for desktop and another design for P cores and E cores in the server (for Intel that would be 4 different core design teams for a single chip), this doesn't seem like a great strategy for a company that is talking about laying off 50% of its engineers IIRC.

Very good point. Desktop is used for more heavy workloads today than in the past. People are likely to browse the internet on their phone. Pretty much anything that is simple to do, and doesn't require lots of screen space to work well is going to be done on a phone app.

Heavy workloads are very often heavily MT with the exception of games.

I would argue that gamers should have a specific core variant that just has some really fast P cores with lots of cache. No need for e Cores for gaming.

Exactly.

For gaming (and likely some other applications) you are correct; however, see my discussion above. The architecture of a new generation specifically targeted for the desktop and laptop? What about DC? What about work station?

Is it even possible for Intel to survive on desktop and laptop sales alone? I don't think so.

DIY PC is a very small market,esp K CPU.
If intel can't offer 20%+ ipc VS zen,V3d is the answer for game pc.

 

[511]

DIY PC is a very small market,esp K CPU.
If intel can't offer 20%+ ipc VS zen,V3d is the answer for game pc.

IPC != gaming performance if the bottleneck is somewhere else if that was the case than Zen5 should be way better than Zen4X3D

 

[MrTeal]

Not sure, but some reports suggest they may have issues anyway:

Things look dire for Intel as Arrow Lake reportedly has instability issues

Stop me if you've heard this one before.

www.xda-developers.com

Arrow Lake instability issues?? RUH ROH!!

Like everyone else, been looking for some news of the new Arrow Lake chips and in my news feed without needing to search found an article about Intel having instability issues early in the game. https://www.xda-developers.com/arrow-lake-reportedly-has-instability-issues/ Not they haven't had...

www.techpowerup.com

That's a lot of layers of obfuscation that eventually points back to MLID and his "sources"

ARL could have issues, but I'd definitely want to see more than his drama of the week video on it.

 

[igor_kavinski]

ARL could have issues, but I'd definitely want to see more than his drama of the week video on it.

One issue that both Level1tech and GN talked about is that the new iGPU (different from previous gens) on Arrow Lake needs to be disabled in BIOS if you have a dGPU otherwise you will get crashes on fresh Windows install (I could be a bit incorrect on the exact details of the issue due to hazy memory). Not sure how Intel validation team missed this gem of a bug.

 

[Hulk]

Both AMD and Intel have cores optimized for ST and MT. The rationale seems to be more for area efficiency rather than energy efficiency as Intel stated when Alder Lake was architecturally released years ago.

AMD's c cores are an area optimized version of its big cores while Intel's area optimized cores are a completely different architecture. But the point is the same, sacrifice ST performance (lower frequency and/or IPC) for more compute in a given area.

AMD has opted to keep SMT so I would call that a "mild" hybrid design. Intel has gone all in. Optimize some P cores for the best throughput possible in terms of IPC and frequency and optimize MT compute/area with E cores.

It took quite a few years for the E cores to go from many called "spam" to Raptor Cove level performance. Part 1 of the promise fulfilled. Part 2, which is the promise of significantly better P core is in doubt currently, I will admit that. I wonder how ARL would have done if it were monolithic?

The 9950X is a beast. I said it when lots of people were down on it at release and I'll say (write) it again now. It's a fabulous design and implimentation. ARL is competitive and one can argue specific use cases where it might be better, but overall, at this point in time I think we have to admit the 9950X is the king of the x86 desktop hill. Overall it wins more benches and is more energy efficient.