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.





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.)

 

[ikjadoon]

Uh ... regarding the Intel power specifications rabbit hole...

Since for Meteor Lake they don't specify base frequency what's the point of specifying base power?

Also can someone please explain "Minimum and Maximum Assured Power?"

I looked it up in Intel's docs and it basically says it's cTDP up/down. I think I'm missing something here?

Edit - Never mind, looks like they are just using new terminology as Max assured power is cTDP up now. Thing is, what's the point of providing these power specs if you don't provide the frequencies to which they refer?

Only turbo has meaning since frequency and power is provided. It almost seems like Intel is hiding something.

Intel frequently hides base frequencies on Ark, but then Intel shows base frequencies to the media. I guess not this year. Example with 13th Gen:







I used to roughly estimate Watts per GHz, but ever since Intel's gone hybrid with E-cores and started hiding easily accessible base frequencies, it's been a crapshoot.

 

[trivik12]

Review from Geekrawan. P-core disappoints but crestmont seems to be good. (Read the entire thread)

https://twitter.com/x/status/1736409726085369861

 

[TESKATLIPOKA]

Review from Geekrawan. P-core disappoints but crestmont seems to be good. (Read the entire thread)

https://twitter.com/x/status/1736409726085369861

Linked the chart, so others don't need to open the link.

 

[SiliconFly]

Not writing MTL off, mainly writing Redwood Cove off, P core is always the most important core, and it's just bad. Zen 5 will be an absolute bloodbath.

Zen 5 is supposed to be Q2 2024 product (some sources say Q3). Arrow Lake is a Q3 2024 product.

Zen 5 will be going against Arrow Lake (not Meteor Lake).

RWC is more of a safe bet for Intel due to MTL's complexity. We all know that the P core performance is abysmal (not something we'd expect in 2023!). Arrow Lake hopefully fixes this issue with LNC.

But I did see some results where MTL's P cores power usage is significantly lower than Raptor Coves P cores. Not sure by how much, but I think it has caught up to competition. Waiting for more results...

 

[Hulk]

Hol up. I am not MLID !

Of course not! I wasn't directing my comments at you. Sorry if there was confusion. I'm not a hater of MLID, in fact I rather enjoy the entertainment of watching/listening to him blab away in his whiney but cheerful tone while that other guy seems to be doing his day job on the computer. I was just doing a little fact check 6 months before a release to show how far off his 6 month prior to release prognostications were.

 

[igor_kavinski]

Curious question: do the AVX-512 units exist in the RWC cores or did Intel finally cut them out completely?

 

[FlameTail]

Of course not! I wasn't directing my comments at you. Sorry if there was confusion.

Take a joke, friend.

 

[Hulk]

Take a joke, friend.

I thought so but wanted to make sure!

 

[FlameTail]

So is Geekerwan the first one to test Meteor Lake with SPEC?

 

[cebri1]

Nice e-core improvement, if the firmware & drivers update improves perf/watt of the gpu and p-cores, MTL will be a very good chip. Specially considering it’s a new node and a chiplet design.

 

[SiliconFly]

Curious question: do the AVX-512 units exist in the RWC cores or did Intel finally cut them out completely?

I remember reading it was fused off in ADL. RPL didn't have it either. So, pretty sure RWC wouldn't have it either.

 

[FlameTail]

Battery Life Test:

Few seconds later, a green bar smoothly popped up:

Needless to say, I chuckled seeing that. This does indeed look ludicrous.

 

[AMDK11]

Curious question: do the AVX-512 units exist in the RWC cores or did Intel finally cut them out completely?

Golden Cove-Raptor Cove and RedwoodCove cores still have 100% AVX512 in their structure.

I doubt Intel will develop two versions of the core with and without AVX512 as it requires more time and cost.

Physically, AVX512 is part of these cores and virtually many structures have been developed with this extension in mind.

 

[Tigerick]

Geez, no wonder Intel don't publish base clock of tCPU. And MTL' tCPU's P-core is divided by two extra big core and four big cores....the density of two types of P-core is different...

 

[H433x0n]

Intel frequently hides base frequencies on Ark, but then Intel shows base frequencies to the media. I guess not this year. Example with 13th Gen:

View attachment 90596


View attachment 90597


I used to roughly estimate Watts per GHz, but ever since Intel's gone hybrid with E-cores and started hiding easily accessible base frequencies, it's been a crapshoot.

I think the reason why is that they legitimately don't know. There's allegedly supposed to be 3 different uCode updates before CES for the various configurations of MTL and a new SoC stepping.

 

[DrMrLordX]

Intel has around 85% market share in client CPUs & Nvidia has around 90% market share in GPUs.

AMD doesn't seem intent on changing that anytime soon. They deliberately cut back on Zen4 client CPU/SoC production to avoid flooding the market in the face of weakening demand. They're more than happy to give that market to Intel for now, and perhaps for awhile yet.

Arrow Lake is a Q3 2024 product.

Ummm what? Since when?

 

[FlameTail]

So Meteor Lake-U series parts only get half the Xe GPU cores?

That's an L.

 

[deasd]

I don't understand. They seem pumping up E core performance to offset the P core regression, which is a bad move. And while even they added more weak cores, efficieny is still behind the competition.

All give me impression they are following every trend like big/little, AI, and imitate Apple, they did all they can do but every metric is underperforming. I'm afraid Intel lost its way.


OTOH what's interesting is, TechEpiphany pointed out the chinese reviewer was cherrypicking:

https://twitter.com/x/status/1736021217428017340

and these leakers turn out to be a joke now as well as MorresLawIsDead.

Intel Meteor Lake to Feature 50% Increase in Efficiency, 2X Faster iGPU

Intel's upcoming Meteor Lake processor family is supposedly looking good with the new performance/efficiency targets. According to the @OneRaichu Twitter account, we have a potential performance estimate for the upcoming SKUs. As the latest information notes, Intel's 14th-generation Meteor Lake...

www.techpowerup.com

According to the @OneRaichu Twitter account, we have a potential performance estimate for the upcoming SKUs. As the latest information notes, Intel's 14th-generation Meteor Lake will feature around a 50% increase in efficiency compared to the 13th-generation Raptor Lake designs.

 

[mikk]

OTOH what's interesting is, TechEpiphany pointed out the chinese reviewer was cherrypicking:

https://twitter.com/x/status/1736021217428017340

I don't see any cherry picking in these games, why? Because he uses League of Legends and Dota? These two games are very very popular in Asia, what is wrong in using them? Why didn't he use the biggest outlier and more of the better games if he is cherry picking? The other games he used are below the average of 10%. In fact Hitman 3, Cyberpunk, Tomb Raider, F1 are mediocre in the Intel slide, why he picked such mediocre titles if he is cherry picking? Furthermore he used 5 of the 8 games also in his 7840HS review from 8 months ago including Dota and League of Legends. It's not like he picked the games just for Meteor Lake. There is no cherry picking involved, Techepiphany is just mad he didn't use the worst games from the Intel slide and that's it.

 

[Meteor Late]

One thing to be careful of is when leakers talk about 50% efficiency increase. They are talking about efficiency related to power consumption at the same performance, which is wayy higher and much easier to achieve than what average people assume efficiency relates to, which is performance at same power.

 

[SiliconFly]

Ummm what? Since when?

Arrow Lake has always been a Q3 2024 product.

 

[João Bortolace]

apparently the sample for thermal testing had two CPU dies... interesting

https://twitter.com/x/status/1736697349014184244

 

[leoneazzurro]

I don't see any cherry picking in these games, why? Because he uses League of Legends and Dota? These two games are very very popular in Asia, what is wrong in using them? Why didn't he use the biggest outlier and more of the better games if he is cherry picking? The other games he used are below the average of 10%. In fact Hitman 3, Cyberpunk, Tomb Raider, F1 are mediocre in the Intel slide, why he picked such mediocre titles if he is cherry picking? Furthermore he used 5 of the 8 games also in his 7840HS review from 8 months ago including Dota and League of Legends. It's not like he picked the games just for Meteor Lake. There is no cherry picking involved, Techepiphany is just mad he didn't use the worst games from the Intel slide and that's it.

These two games are so light that the test is practically always CPU bound, so for testing the GPU these are not the best examples. As a general review, these are OK, one must keep present that on more GPU demanding titles the situation may be quite different.

 

[igor_kavinski]

P-core is divided by two extra big core and four big cores....the density of two types of P-core is different...

Uh WHAT???

Why isn't Intel mentioning this in their MTL launch slides? Why do they have dissimilar P-core clusters? Does the first cluster have more cache than the second cluster or just that the first cluster has more cache available to share between its two P-cores? This is getting ridiculous. So now there are multiple levels of cores in an Intel CPU. Level 1 and Level 2 P-cores, then E-cores and finally LP E-cores.

 

[DrMrLordX]

Arrow Lake has always been a Q3 2024 product.

No, it's always been a Q4 2024 product, if you're being optimistic.