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.)

 

[SiliconFly]

Semiwiki article is from june 2022 and is about expectations, not actual silicon.

Which part of the article don't you understand? These article is based on published numbers from the respective foundries & not expectations. It's the reality.

 

[TESKATLIPOKA]

Even if the designs are different MTL has 6P, and the 8E should more than compensate for the 7840 using 8P, MTL is supposed to be on par with a theorical 10P CPU.

On par with a theoretical 10P Intel CPU.

At the very least it could be clocked such that it would have vastly better perf/Watt at same perfs assuming an equal process is used, and since this didnt happen it means that the culprit is the process, Intel s P core could be less efficient than a Zen 4 core but not to this extent.

You would need the same core to be made using N5 and Intel 4, then you can be sure what is actually the problem.

We will need to wait for a few months, to make a valid conclusion how good MTL really is.

 

[SiliconFly]

If Intel 4 was to be better than TSMC s N5 as said in the article then a 6 + 8 MTL would eat a 7840 for breakfeast in anything multithreaded perf/watt wise, obviously the expectations were above the final result.

One is not gonna eat the other for breakfast just due to nodes alone. The CPU architecture also matters a lot. Comparing node characteristics is pretty much straightforward. But comparing processors with different architectures is apples and oranges. RWC is nothing but Raptor Cove (RPC) which is nothing but a minor tweak of GLC and these architectures are basically designed for performance over power efficiency.

Can't expect much power efficiency for MTL due to RWC alone. For MTL's overall power-efficiency, RWC needs to work in tandem with LPE cores to achieve higher power efficiency. Production MTL laptops with updated pcode should provide a clearer picture I think.

 

[tamz_msc]

We got two "review" laptops - one from Asus and the other from Acer, who are known for their spotty BIOS support. Even with sub-par BIOS, we got pretty impressive results for MTL on Windows.

At least wait till CES with the likes of Dell and Lenovo releasing their products with MTL. A good Inspiron or IdeaPad would be representative of what MTL is capable of.

 

[TESKATLIPOKA]

We got two "review" laptops - one from Asus and the other from Acer, who are known for their spotty BIOS support. Even with sub-par BIOS, we got pretty impressive results for MTL on Windows.

At least wait till CES with the likes of Dell and Lenovo releasing their products with MTL. A good Inspiron or IdeaPad would be representative of what MTL is capable of.

There will be a Lenovo IdeaPad Pro 5 14IMH9 and Lenovo IdeaPad Pro 5 16IMH9 with MTL.
Both of them should have a 2 fan cooling system. Not sure about the bigger sibling, but I have the 14-inch one with PHX and It can cool 60W power limit without a problem, so I think MTL will have a comparable power limit or a bit higher.

 

[FlameTail]

TSMC CEO said N3P is competitive with Intel 18A.

Intel CEO said 18A is competitive with TSMC N2.

It's wild that a node without GAAFET or BSPD is competitive with Intel 18A- which has both of them.

If that is indeed true, TSMC I truly on another level.

 

[H433x0n]

What I just heard is the new BIOS is to lower the voltage and boost higher, considering ASUS has bad track record that set voltage too high and burn Alderlake/Zen4, I guess it's just only ASUS has this problem, that being said it's not Intel's fault, and those reviewers did nothing wrong.

I don’t think it has anything to do with vf curve. I think it’s either bandwidth between tiles or an optimization of when to shut off LP E-Cores.

 

[DAPUNISHER]

Highly doubt It. I don't think they have enough volume for that.

That's a shame. The ARC iGPU does really well in the Phoronix suite. Link for anyone that missed it https://www.phoronix.com/review/meteor-lake-arc-graphics

With all of the improvements coming via software updates Meteor Lake will keep getting better too.

 

[FlameTail]

Zen 5
• Major architecture upgrade
• Minor node upgrade (N5P -> N4P)

Arrow Lake
• Major architecture upgrade (Lion Cove/Royal Core)
• Major node upgrade (Intel 7 -> 4 -> 20A)

Arrow Lake is getting 2 Major node jumps from Intel 7 (RPL) to Intel 20A- with GAAFETs and PowerVia to boot.

I was thinking Arrow Lake might beat Zen 5's performance but lose out on efficiency. But then the realisation that Arrow Lake is getting a jump of 2 Major nodes, completely changed my perspective. Now I am thinking Arrow Lake has the potential to not just match Zen 5 in performance/efficiency, but also beat it in those aspects.

 

[igor_kavinski]

also beat it in those aspects.

That would be a big deal if it happens. Not very optimistic about it. Intel 7 is a very mature node while 20A would be refined just enough to make ARL CPUs in volume. It would be surprising if 20A hits 5.5 GHz on the top SKU.

 

[Henry swagger]

Zen 5
• Major architecture upgrade
• Minor node upgrade (N5P -> N4P)

Arrow Lake
• Major architecture upgrade (Lion Cove/Royal Core)
• Major node upgrade (Intel 7 -> 4 -> 20A)

Arrow Lake is getting 2 Major node jumps from Intel 7 (RPL) to Intel 20A- with GAAFETs and PowerVia to boot.

I was thinking Arrow Lake might beat Zen 5's performance but lose out on efficiency. But then the realisation that Arrow Lake is getting a jump of 2 Major nodes, completely changed my perspective. Now I am thinking Arrow Lake has the potential to not just match Zen 5 in performance/efficiency, but also beat it in those aspects.

Will depend on clock speeds that will decide

 

[igor_kavinski]

Meteor Lake EPP Tuning For Greater Performance Or Power Efficiency With Intel Ultra Core 7 - Phoronix

www.phoronix.com

If wanting to extend an Intel Meteor Lake laptop's battery life, the testing on this Acer laptop showed the EPP power-saver reducing the performance by about a quarter of the original performance. While the performance was 73% the original (balanced) default, the power consumption was on average at 63% the balanced rate.

Lenovo ThinkPad P14s Gen 4 w/ AMD Ryzen 7 PRO 7840U Running Nicely On Linux - Phoronix

www.phoronix.com

So 95% the performance at 82% the power use for those wanting to maximize the power efficiency of the Lenovo ThinkPad P14s Gen 4.

73% perf at 63% lower power use for MTL while 7840U gets 95% perf at 82% of normal power use. I don't know about anyone else here but 7840U seems like a perf/watt beast to me that barely loses any performance when in power saving mode.

 

[Hitman928]

First English language review of a retail unit I've seen. PCWorld reviewed the ASUS Zenbook. The details are scant but their Cinebench r23 score roughly matches the Chinese reviewer of the same unit, both of which are in line with the initial Cinebench scores rather than the reported updated firmware scores. So for at least this unit, either the retail bios doesn't have the reported updated firmware, there is something else going on with the updated firmware scores from others, or there is some kind of serious cooling or config issue with the ASUS retail model.

We test the first Meteor Lake computer from Asus

The new Asus Zenbook 14 OLED has solid battery life and a gorgeous OLED screen. That said, it's not much faster than its predecessors even with the high-efficiency Intel Core Ultra processor.

www.pcworld.com

 

[SiliconFly]

Zen 5
• Major architecture upgrade
• Minor node upgrade (N5P -> N4P)

Arrow Lake
• Major architecture upgrade (Lion Cove/Royal Core)
• Major node upgrade (Intel 7 -> 4 -> 20A)

Arrow Lake is getting 2 Major node jumps from Intel 7 (RPL) to Intel 20A- with GAAFETs and PowerVia to boot.

I was thinking Arrow Lake might beat Zen 5's performance but lose out on efficiency. But then the realisation that Arrow Lake is getting a jump of 2 Major nodes, completely changed my perspective. Now I am thinking Arrow Lake has the potential to not just match Zen 5 in performance/efficiency, but also beat it in those aspects.

Well, Intel 7 being a very mature node (10+++++ or 10 followed by literally 5 pluses) has a higher Fmax. It clocks really really high (like > 6 GHz).

20A being a new node might not clock well and might actually have Fmax regression. Imho, 20A may not clock well above 5 GHz since it's the first iteration (maybe 5.1 or 5.2 at best considering history). That is, given the exact same architecture, a CPU going from Intel 7 directly to 20A might experience around 15% to 20% performance regression. Sad but true.

But 20A offers very high PPW over Intel 7. So, given the same architecture, a CPU going from Intel 7 directly to 20A will get a solid efficiency boost of up to 60% due to the node jump alone. (source: AnandTech)

The performance uplift we can look forward in ARL is gonna come from the new LNC architecture rather than the node jump. In fact, if 20A doesn't clock well, then LNC has to bring in a ton of performance uplift just to compensate for the loss in performance due to lower Fmax.

Bottom line, in desktops, it looks very likely that ARL is set to comfortably beat the previous gen (RPL-R) and the competition (Zen 5) in power efficiency by a wide margin. But, beating the previous gen or the competition in performance is an uphill task. And considering RPL-R & Zen 5 are both supposed to be top performers, ARL beating them in performance in desktops with lower boost clocks is a very big question. LNC can't be THAT good!

But on the mobile side, where ~5 GHz clocks are the norm, ARL might kick ass! As there won't be any clock regression there. Might beat the previous gen and the competition in power efficiency by a wide margin and in performance by a decent margin.

 

[Markfw]

Will depend on clock speeds that will decide

I think there are lot more factors that will decide. I won't venture a guess this early, but Intels track record as of late has not been great.

 

[adroc_thurston]

Now I am thinking Arrow Lake has the potential to not just match Zen 5 in performance/efficiency, but also beat it in those aspects.

It's not even close lol.
Please forward your hopium to Lunar Lake.

 

[mikk]

First English language review of a retail unit I've seen. PCWorld reviewed the ASUS Zenbook. The details are scant but their Cinebench r23 score roughly matches the Chinese reviewer of the same unit, both of which are in line with the initial Cinebench scores rather than the reported updated firmware scores. So for at least this unit, either the retail bios doesn't have the reported updated firmware, there is something else going on with the updated firmware scores from others, or there is some kind of serious cooling or config issue with the ASUS retail model.

We test the first Meteor Lake computer from Asus

The new Asus Zenbook 14 OLED has solid battery life and a gorgeous OLED screen. That said, it's not much faster than its predecessors even with the high-efficiency Intel Core Ultra processor.

www.pcworld.com

This is not a retail unit, it's a pre-sample from Asus posted two weeks ago on pcforalla.se, maybe read carefully. Also it's not a given a retail performs better, who says this? It's down to the bios/firmware which might or might not unlock a better performance.

 

[TESKATLIPOKA]

Zen 5
• Major architecture upgrade
• Minor node upgrade (N5P -> N4P)

Arrow Lake
• Major architecture upgrade (Lion Cove/Royal Core)
• Major node upgrade (Intel 7 -> 4 -> 20A)

Arrow Lake is getting 2 Major node jumps from Intel 7 (RPL) to Intel 20A- with GAAFETs and PowerVia to boot.

I was thinking Arrow Lake might beat Zen 5's performance but lose out on efficiency. But then the realisation that Arrow Lake is getting a jump of 2 Major nodes, completely changed my perspective. Now I am thinking Arrow Lake has the potential to not just match Zen 5 in performance/efficiency, but also beat it in those aspects.

If the core count doesn't change on both sides, then I wouldn't be particularly optimistic about ARL.
It's very likely that we will see some clock regression. This won't be as noticeable at full load, because power will be the limiting factor for frequency.
But expecting ARL to have higher IPC gain than Zen5 is rather unrealistic, so I have to ask why It should beat Zen5 not just in perf/W but also performance?

 

[FlameTail]

And expecting ARL to have higher IPC gain than Zen5 is rather unrealistic

Why not? ARL has Lion Cove, major architectural redesign with no Hypethreading and Royal Core Project stuff.

 

[adroc_thurston]

Why not?

Because IDC is washed up and no good.
really simple!

major architectural redesign with no Hypethreading and Royal Core Project stuff.

It's literally just more bloat.

 

[TESKATLIPOKA]

Why not? ARL has Lion Cove, major architectural redesign with no Hypethreading and Royal Core Project stuff.

Who said ARL won't have Hyperthreading? Just this alone would shave of 10-15% of MT performance.
There is no info about ARL, what little was leaked actually talks about only single digit IPC gain, If I remember correctly.

 

[adroc_thurston]

Who said ARL won't have Hyperthreading?

Well it just doesn't.
No LNC part has it.

There is no info about ARL, what little was leaked actually talks about only single digit IPC gain, If I remember correctly.

Performance, not IPC.
Again, LNC issue is it being more of the same.
Which is bloat.

 

[SiliconFly]

It's literally just more bloat.

LNC has been in development for a very long time. Nothing much is known about the actual architecture.

It may turn out to be a dud. Or it may turn out to be the best in the world!

 

[TESKATLIPOKA]

Well it just doesn't.
No LNC part has it.

Then I have to wonder how thy plan to compensate. Most likely more E-cores.

LNC has been in development for a very long time. Nothing much is known about the actual architecture.

It may turn out to be a dud. Or it may turn out to be the best in the world!

I am kinda worried about It being in development for a very long time. In most cases that doesn't end in a very good product.

Not like It really matters. If It will be missing performance, then just lower the price and you have a competitive product.

 

[Hitman928]

This is not a retail unit, it's a pre-sample from Asus posted two weeks ago on pcforalla.se, maybe read carefully. Also it's not a given a retail performs better, who says this? It's down to the bios/firmware which might or might not unlock a better performance.

Thanks for the correction but there's nothing on the site that indicates it was from 2 weeks ago. They do give credit to pcforalla but they date it as today on the English version.