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.





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.)

 

[dullard]

If "desktop" means Meteor Lake mobile parts in small form computers then I don't count that as desktop.
I'm obviously missing something here.
My conundrum here is how do do Meteor Lake and Arrow Lake coexist on the desktop? Wouldn't they both be released at about the same time? Arrow would be built on a newer node and have a better P core. Why Meteor desktop at all?

Reading between the lines it looks like Meteor Lake desktop will be the lower power, lower performance desktop chips. Arrow Lake desktop looks like it will be the higher powered K overclockable chips. They would be in totally different market segments even if they are both desktop chips (It isn't like i3 and i9 chips are competing for the same customers). I have not seen any confirmation of that, but all the tea-leaves seem to be pointing that way.

Why do it? De-risking. Intel has already announced that they will be overlapping products for that very reason. They were burned badly by putting all of their eggs in one basket that failed to be work well (10 nm). They are very deliberately going down multiple simultaneous paths to de-risk and avoid that situation from happening again. The counter argument is that going down multiple overlapping paths is costly. That is true, but only to a certain extent--Intel does not seem to be duplicating every die combination on different nodes.

Take Alder Lake for example, there were 4 different Alder Lake dies. (1) 8P+8E desktop, (2) 6P desktop, (3) 6P+8E mobile, and (4) 2P+8E mobile. https://en.wikipedia.org/wiki/Alder_Lake#Dies It does not appear that Intel is making 4 Meteor Lake dies and 4 Arrow Lake dies. But instead, Intel is probably making a smaller number of dies on each node. That way, there isn't as much duplicated cost. But they still get to de-risk in case one node is delayed or is an outright failure. They will still have desktop chips to sell in that case.

So think of Meteor Lake desktop and Arrow Lake desktop as the two different desktop dies that Alder Lake had. A lower powered, lower performance one. And a high performance one. Plus, it gives Intel a great option to include different mixtures of the other chiplets. Maybe the K high-powered chips don't need as much iGPU performance as those would often be paired with a discrete GPU.

 

[Hulk]

Reading between the lines it looks like Meteor Lake desktop will be the lower power, lower performance desktop chips. Arrow Lake desktop looks like it will be the power powered K overclockable chips. They would be in totally different market segments even if they are both desktop chips (It isn't like I3 and I9 chips are competing for the same customers). I have not seen any confirmation of that, but all the tea-leaves seem to be pointing that way.

Why do it? De-risking. Intel has already announced that they will be overlapping products for that very reason. They were burned badly by putting all of their eggs in one basket that failed to be work well (10 nm). They are very deliberately going down multiple simultaneous paths to de-risk and avoid that situation from happening again. The counter argument is that going down multiple overlapping paths is costly. That is true, but only to a certain extent--Intel does not seem to be duplicating every die combination on different nodes.

Take Alder Lake for example, there were 4 different Alder Lake dies. (1) 8P+8E desktop, (2) 6P desktop, (3) 6P+8E mobile, and (4) 2P+8E mobile. https://en.wikipedia.org/wiki/Alder_Lake#Dies It does not appear that Intel is making 4 Meteor Lake dies and 4 Arrow Lake dies. But instead, Intel is probably making a smaller number of dies on each node. That way, there isn't as much duplicated cost. But they still get to de-risk in case one node is delayed or is an outright failure. They will still have desktop chips to sell in that case.

So think of Meteor Lake desktop and Arrow Lake desktop as the two different desktop dies that Alder Lake had. A lower powered, lower performance one. And a high performance one. Plus, it gives Intel a great option to include different mixtures of the other chiplets. Maybe the K high-powered chips don't need as much iGPU performance as those would often be paired with a discrete GPU.

Okay, that makes sense but there are two things I'm wondering. First, if the Arrow Lake P core has better IPC than Meteor's P then that will create some confusion in the stack won't it?
Second, aren't the chips with defects, which make cores inoperable, and don't clock as high the ones that end up in the value segment?
Actually three questions. I can't think of a time when Intel had two different cores in the same desktop generation? Perhaps with some Core and Netburst during the transition?
Just seems like a convoluted strategy.

One outcome I could see occurring is if Arrow Lake on 20A can't clock high enough to compete/beat Raptor Refresh and ends up mobile. Meteor on a Refresh could perhaps move into desktop then?

I don't know, the future, cloudy it is.

 

[Abwx]

This right here, is not the default of what perf/watt means.
When Intel says that they gain 20% perf/watt from Intel 4, it means they gained 20% more performance iso power vs Intel 7. This is shown in this graph:
View attachment 86187
Notice how they equalized power, not performance for their perf/watt claim.
"perf/watt iso perf" would be watt/perf. Phoenix uses 45watts/16000 points, RPL uses 65 watts/16000 points. RPL uses 44% more watts per point at 16,000 points.
The reason why the denominator is usually equalized is because the fraction sign is understood as "per", for example 60 miles per hour.

You would need to know the form of the power/frequency curve for literally everything. Even perf/watt iso perf. For example, at 16000 points, RPL uses 44% more power, but at 12,000 points, RPL uses ~53% more energy.

A 6 + 8 13700H use close to 89W to score 16000 and 51W to score about 13300 :

https://www.notebookcheck.net/Dell-Inspiron-16-Plus-7630-laptop-review-60-W-GeForce-RTX-4060-isn-t-so-bad.743747.0.html

So much for your 16k/ 65W, reality is 30-37% more power than your flawed assumptions.

Here the 13700H page at NBC, you can see the powers and scores in CB R23, none is even remotely close of your statements :

https://www.notebookcheck.net/Intel-Core-i7-13700H-Processor-Benchmarks-and-Specs.677405.0.html

So yes, MTL gaining 25% better perf/watt over RPL should make it enough to roughly match Phoenix, using that notebook check graph provided, at least for the power levels described on the graph. And yes, we should be looking at improvements over a 6+8 RPL.

21.5% better perf at isopower is at 2.1GHz, as frequency increase the gain is shrinking, there s an article at AT :

Intel 4 Process Node In Detail: 2x Density Scaling, 20% Improved Performance

www.anandtech.com

 

[dullard]

Okay, that makes sense but there are two things I'm wondering. First, if the Arrow Lake P core has better IPC than Meteor's P then that will create some confusion in the stack won't it?
Second, aren't the chips with defects, which make cores inoperable, and don't clock as high the ones that end up in the value segment?
Actually three questions. I can't think of a time when Intel had two different cores in the same desktop generation? Perhaps with some Core and Netburst during the transition?
Just seems like a convoluted strategy.

One outcome I could see occurring is if Arrow Lake on 20A can't clock high enough to compete/beat Raptor Refresh and ends up mobile. Meteor on a Refresh could perhaps move into desktop then?

I don't know, the future, cloudy it is.

I have a vague recollection of blended nodes. So, I did a quick search and came up with this Anandtech speculation from a few years ago:

Ashraf also mentioned words from Dr. Venkata ‘Murthy’ Renduchintala, VP and GM of Client and IoT:

Murthy referred to it at the event, process tech use will be ‘fluid’ based on segment’.

If one read too much into this, we may start seeing a blend of process nodes for different segments at the same time for different areas of the market...We may get to a point where Intel's 'Gen' naming scheme for its CPUs covers two or more process node variants.

Intel Confirms 8th Gen Core on 14nm, Data Center First to New Nodes

www.anandtech.com

Others do not think at the same level about chips as we do. Most just go to a store/online and buy the gaming laptop. Or the cheap desktop. Or the best they can afford. Most don't have a clue or even care what node the chip is made on. I don't think hardly any customer will be confused simply because they won't even know there is a difference. At most, they MIGHT know that i9 > i3. Or they MIGHT know that 13900 > 13100. And that is only if marketing does its job. They certainly won't know if one is on Intel 4 and another is on Intel 20A (and they won't know which is better).

 

[H433x0n]

21.5% better perf at isopower is at 2.1GHz, as frequency increase the gain is shrinking, there s an article at AT :

Intel 4 Process Node In Detail: 2x Density Scaling, 20% Improved Performance

www.anandtech.com

At higher frequency 13700H is only <10% behind 7940HS anyway. Its at the lower voltage regions where it really gets gapped.

 

[Abwx]

At higher frequency 13700H is only <10% behind 7940HS anyway. Its at the lower voltage regions where it really gets gapped.

It is in the regions that matters, i dont think that people are aware that by buying Intel based laptops they are buying mobile furnaces with a CPU like the 13700H that can be set higher than 100W PL2 in some configurations, the NBC link i posted is telling.

 

[H433x0n]

It is in the regions that matters, i dont think that people are aware that by buying Intel based laptops they are buying mobile furnaces with a CPU like the 13700H that can be set higher than 100W PL2 in some configurations, the NBC link i posted is telling.

I honestly don’t think anybody would notice.

Did anybody consider Zen 3 mobile processors furnaces? I don’t think they did. So why would a 13700H laptop be a “furnace” when it has better perf/watt than a laptop equipped with a 7735HS?

In a vast majority of cases people are upgrading laptops that are 3+ years old and from that perspective it’s a massive step forward.

 

[Abwx]

I honestly don’t think anybody would notice.

Did anybody consider Zen 3 mobile processors furnaces? I don’t think they did. So why would a 13700H laptop be a “furnace” when it has better perf/watt than a laptop equipped with a 7735HS?

In a vast majority of cases people are upgrading laptops that are 3+ years old and from that perspective it’s a massive step forward.

A 6nm fabbed 6850U score 10 000 pts at 30W, a 13700H score 12 000 pts at 45W, so at 30W it would barely reach the 10k mark, actually even at 6 + 8 RPL mobile is somewhat less efficient than a 8C Rembrandt, what about if we pit a 6 + 4 instead, wich is the real comparison..?..

Anyway that s prove that Intel marketing is much more efficient than their CPUs, since they manage to make their products way overestimated even by alleged technical forumers...

https://www.notebookcheck.net/Lenovo-ThinkPad-T14s-G3-AMD-laptop-review-Quiet-and-efficient-workhorse-with-Ryzen-power.682906.0.html

https://www.notebookcheck.net/Schenker-Vision-16-2023-Review-16-inch-lightweight-with-many-ports.738862.0.html

 

[igor_kavinski]

https://www.notebookcheck.net/Lenovo-ThinkPad-T14s-G3-AMD-laptop-review-Quiet-and-efficient-workhorse-with-Ryzen-power.682906.0.html

https://www.notebookcheck.net/Schenker-Vision-16-2023-Review-16-inch-lightweight-with-many-ports.738862.0.html

Very interesting. The Thinkpad is hotter on the upper side but maintains a relatively cooler bottom side. Better thermal design than the Schenker.

 

[Abwx]

Very interesting. The Thinkpad is hotter on the upper side but maintains a relatively cooler bottom side. Better thermal design than the Schenker.

We cant really compare, the 6850U is a 14" while the 13700H is 16", in principle the latter should perform thermaly as well as the former given the bigger size, but then there s the design and fan speeds to consider, the Thinkpad noise is in a 23-28dB range and the Schenker within 25-50dB.

 

[SiliconFly]

It is in the regions that matters, i dont think that people are aware that by buying Intel based laptops they are buying mobile furnaces with a CPU like the 13700H that can be set higher than 100W PL2 in some configurations, the NBC link i posted is telling.

Thats the dumbest thing i've heard. I have a 13700HX laptop and it doesn't run hot when compared to other laptops. Afaik, Intel laptops are better value when compared to cheap ryzen laptops.

 

[cebri1]

"We can't really compare" -> We only compare when it suits us.

I'm eager to get the first reviews, hopefully early December when I hope Intel will release some official figures and some test laptops will be sent to reviewers.

 

[Abwx]

Thats the dumbest thing i've heard. I have a 13700HX laptop and it doesn't run hot when compared to other laptops. Afaik, Intel laptops are better value when compared to cheap ryzen laptops.

Dumbest thing is to use one own exemple as a general rule, look at CB R23 MT scores, there s the TDP of the 13700H laptops under test :

https://www.notebookcheck.net/Intel-Core-i7-13700H-Processor-Benchmarks-and-Specs.677405.0.html

Most are 50W at least with PL2s way above this value, some are even at 100W PL1, that being said surely that at those powers a laptop can run cool enough if the fan is at full tilt..

As for cheap AMD Ryzen that just plain deffamation, the market is full of half baked Intel laptops, or rather furnaces, find me a single Intel laptop at this price that has as well balanced perfs and caracteristics :

https://www.notebookcheck.net/HP-EliteBook-845-G10-review-Business-laptop-impresses-with-AMD-Ryzen-7-7840U-and-unbeatable-price.754141.0.html

 

[Abwx]

"We can't really compare" -> We only compare when it suits us.

I'm eager to get the first reviews, hopefully early December when I hope Intel will release some official figures and some test laptops will be sent to reviewers.

I said that thermals were not comparable because the 13700H is at 45W and the 6850U at 30W, look like you re so Intel addicted that you dont even realise what it is about, butknow if you want we can make the comparison and state that the 6850U is much better thermaly designed, that suit your stance now..?.

 

[cebri1]

I said that thermals were not comparable because the 13700H is at 45W and the 6850U at 30W, look like you re so Intel addicted that you dont even realise what it is about, butknow if you want we can make the comparison and state that the 6850U is much better thermaly designed, that suit your stance now..?.

I have an AMD desktop. But I'm not a white knight nor a doomer for certain companies. I like to discuss Intel because they are in turn around mode which is way more interesting than analyzing NVIDIA dominance of the AI space.

But you are cherry picking data, I will post a more detailed analysis of CB20 MT results, because there is a high degree of variability, precisely because some chips are thermally throttled. And let's not forget this is "leaked" data, we don't know conditions or even if the results will mimic real world performance considering there are significant changes.

 

[Abwx]

But you are cherry picking data, I will post a more detailed analysis of CB20 MT results, because there is a high degree of variability, precisely because some chips are thermally throttled. And let's not forget this is "leaked" data, we don't know conditions or even if the results will mimic real world performance considering there are significant changes.

NBC measure power comsumption of the whole laptops with an external screen, there s curves that show the power evolution, beside they test CB R15 for several loops and report the score, it s clearly visible how much the CPU lose perf from PL2 to PL1, so thermal throttling is under control when it comes to measurements.

Truth is that people are unaware of Intel s CPUs much lower efficency, so when real numbers hit they have a hard time understanding the difference between the marketing driven distorted brand perception and actual perfs...

 

[cebri1]

NBC measure power comsumption of the whole laptops with an external screen, there s curves that show the power evolution, beside they test CB R15 for several loops and report the score, it s clearly visible how much the CPU lose perf from PL2 to PL1, so thermal throttling is under control when it comes to measurements.

Truth is that people are unaware of Intel s CPUs much lower efficency, so when real numbers hit they have a hard time understanding the difference between the marketing driven distorted brand perception and actual perfs...

You keep making wrong assumptions, Intel CPUs low efficiency at the higher part of curve is known to anyone that has bothered to check. We are not judging that (or at least me) but what's the improvement from Intel 7 to Intel 4, as Redwood Cove seems to be very similar to Raptor Cove and the major changes are in the e-core and SOC.

 

[Abwx]

You keep making wrong assumptions, Intel CPUs low efficiency is known to anyone that has bothered to check. We are not judging that (or at least me) but what's the improvement from Intel 7 to Intel 4, as Redwood Cove seems to be very similar to Raptor Cove and the major changes are in the e-core and SOC.

It is not known that much, you can see from some people here who overestimate Intel s efficency, and yet, we are supposed to be in a technical forum, indeed all this discussion started because a member precisely talked of a score at 65W while it was performed at 80-89W...

As for future Intel products MLID leak state a 7300 score in CB R20 for a 6 + 8 MTL, allegedly at 100W.

7300 pts in R20 amount to about 18700-19000 pts in CB R23, a comparable RPL, that is a 6 + 8 13700H, does 16000 pts at about 89W, so that s about the 18% better perf/isowatt that is brought by Intel 4.

According to AT Intel s 4 21.5% better perf/isowatt is at 2.1GHz, as frequency increase the percentage shrink, so a 18% figure above 3GHz is realistic.

 

[cebri1]

It is not known that much, you can see from some people here who overestimate Intel s efficency, and yet, we are supposed to be in a technical forum, indeed all this discussion started because a member precisely talked of a score at 65W while it was performed at 80-89W...

As for future Intel products MLID leak state a 7300 score in CB R20 for a 6 + 8 MTL, allegedly at 100W.

7300 pts in R20 amount to about 18700-19000 pts in CB R23, a comparable RPL, that is a 6 + 8 13700H, does 16000 pts at about 89W, so that s about the 18% better perf/isowatt that is brought by Intel 4.

According to AT Intel s 4 21.5% better perf/isowatt is at 2.1GHz, as frequency increase the percentage shrink, so a 18% figure above 3GHz is realistic.

Point 1: I'm now realizing that I misunderstood your first post. I apologize.

Point 2: And that's a realistic expectation so I don't know why are some people talking about doom and gloom regarding the MLID video > "power draw is awful" ... compared to RPL-H? No, compared to AMD, for sure. But I would like to wait as well for the lower powered MTL products, both ADL and RPL efficiency improved significantly at the lower part of the curve and I expect MTL to be similar. For high performance Intel will probably release RPL refresh for the HX product line (or whatever is called now).

Point 3: mega crude comparison, as some results do not add up (notebookcheck is not that reliable imo), but there is an obvious trend.

 

[SiliconFly]

...buying Intel based laptops they are buying mobile furnaces...

This is just plain and pure defamation. Take your AMD worshiping to an AMD thread. This is an Intel thread.

...are even at 100W PL1, that being said surely that at those powers a laptop can run cool enough if the fan is at full tilt..

13700H is designed to run at 45W PL1 in a laptop. If anybody is insane enough to override that and set it to 100W, the laptop's cooling solution won't be able to handle the heat and will cause a CPU meltdown unless they have a liquid cooled laptop. Not practical.

...the market is full of half baked Intel laptops, or rather furnaces...

The build quality of Intel laptops are far superior to AMD laptops in general. Thats why people buy way more Intel laptop the cheap AMD ones.

the 6850U is much better thermaly designed, that suit your stance now..?.

You think anyone's gonna buy a 6850U over a 6+8 13700H? In ur dreams...

But you are cherry picking data...

That he does usually. Sad but true.

...and the major changes are in the e-core and SOC.

We can't speak to people who pretend to be deaf. Efficiency optimized Intel 4 HP cell library, LP E-cores, DLVR, independent power-down of tiles, etc. But he doesn't want to listen!

According to AT Intel s 4 21.5% better perf/isowatt...

MTL expected 2X power-efficiency improvement is not only because of the node jump but also because of it's new superior hyper-efficient architecture. MTL final is expected to beat the competition in perf-per-watt by a mile.

As for future Intel products MLID leak state a 7300 score in CB R20 for a 6 + 8 MTL, allegedly at 100W.

And you still believe in MLID. Oh my! That guy is a well-known Intel hater who cherry picks data and highlights unreliable leaks.

 

[Abwx]

13700H is designed to run at 45W PL1 in a laptop. If anybody is insane enough to override that and set it to 100W, the laptop's cooling solution won't be able to handle the heat and will cause a CPU meltdown unless they have a liquid cooled laptop. Not practical.

And yet most have PL1 above 45W.

The build quality of Intel laptops are far superior to AMD laptops in general. Thats why people buy way more Intel laptop the cheap AMD ones.


You think anyone's gonna buy a 6850U over a 6+8 13700H? In ur dreams...

Any day, better something that run cool with good perfs.

That he does usually. Sad but true.


We can't speak to people who pretend to be deaf. Efficiency optimized Intel 4 HP cell library, LP E-cores, DLVR, independent power-down of tiles, etc. But he doesn't want to listen!


MTL expected 2X power-efficiency improvement is not only because of the node jump but also because of it's new superior hyper-efficient architecture.


And you still believe in MLID. Oh my! That guy is a well-known Intel hater who cherry picks data and highlights unreliable leaks.

His numbers are in line with Intel s statement about Intel 4, 21.5% at 2.1Ghz and the number shrinking as frequency increase, that s in their slides, rest is bla bla...

 

[cebri1]

MTL expected 2X power-efficiency improvement is not only because of the node jump but also because of it's new superior hyper-efficient architecture. MTL final is expected to beat the competition in perf-per-watt by a mile.


And you still believe in MLID. Oh my! That guy is a well-known Intel hater who cherry picks data and highlights unreliable leaks.

I do believe that MTL will be more efficient in day to day workloads vs synthetic benchmarks, just due to the new tile approach and new thread director. But I dont expect 2x efficiency, although at the lower levels of the power curve I think MTL will be better than most people expect.

 

[SiliconFly]

I do believe that MTL will be more efficient in day to day workloads vs synthetic benchmarks, just due to the new tile approach and new thread director. But I dont expect 2x efficiency, although at the lower levels of the power curve I think MTL will be better than most people expect.

Exactly. I think when Intel says 2X, I think it's under normal circumstances. And thats what really matters. No one's going to run synthetic benchmarks 24/7 all round the year in their MTL laptops imho.

 

[Abwx]

Point 1: I'm now realizing that I misunderstood your first post. I apologize.

Point 2: And that's a realistic expectation so I don't know why are some people talking about doom and gloom regarding the MLID video > "power draw is awful" ... compared to RPL-H? No, compared to AMD, for sure. But I would like to wait as well for the lower powered MTL products, both ADL and RPL efficiency improved significantly at the lower part of the curve and I expect MTL to be similar. For high performance Intel will probably release RPL refresh for the HX product line (or whatever is called now).

Point 3: mega crude comparison, as some results do not add up (notebookcheck is not that reliable imo), but there is an obvious trend.

View attachment 86241

His numbers are not far from what Intel published for the process used, and that s a pre production chip, it can easily be set at say 5% higher voltage for the tests and consume 100W instead of say 89W for the final revision.

That being said i dont believe that he s capable to do the maths required to get to these conclusions since he has no understanding of transistors physics, so he had forcibly access to some confidential infos that he understood more or less accurately.

 

[SiliconFly]

His (MLID) numbers are in line with Intel s statement about Intel 4, 21.5% at 2.1Ghz...

His numbers are not far from what Intel published for the process used...

he has no understanding of transistors physics...

Still quoting the baboon? He's an Intel-hater with poor understanding of tech and cherry-picks data.

I'll repeat. MTL's so-called 2X power efficiency is due to its new superior hyper-efficient architecture and not the node jump alone.

Just fyi, do you know that MTL's NOC Fabric is far superior to AMD's Infinity fabric? It can deliver up to 4X the bandwidth under certain conditions compared to competition. Also very efficient. MTL's efficiency comes from lots of under-the-hood improvements. Think you missed Intel Innovation.


Name calling and insults are still not allowed.


esquared
Anandtech Forum Director