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.)

 

[cebri1]

Intel's slapped a 16GB dual-channel minimum entry requirement on Core Ultra laptops if you want top gaming performance

Intel's slapped a 16GB dual-channel minimum entry requirement on Core Ultra laptops if you want top gaming performance

To get an 'Arc Graphics' sticker, your Core Ultra laptop has to be dual-channel with 16GB of memory.

www.pcgamer.com

Intel slapped a

This is not news. With current Xe graphics laptops is exactly the same. Xe Graphics with dual slot memory, UHD graphics with one slot.

edit: eg. https://support.hp.com/us-en/document/c06978058

Intel Iris Xe Graphics
note: Intel Iris Xe Graphics capabilities require system to be configured with Intel Core i5 or i7 processors and dual channel memory.

 

[FlameTail]

Thank god. We need 8GB memory to die in 2023. Memory is not that expensive. BTW are there any laptops with LPDDR5x7466. That is the fastest supported memory right?

On Early issues even Apple iphone has had early issues which was fixed with further software updates. I think Intel was fixated on getting it out before end of the year. Probably needed more QA time for packaging team. its something unprecedented(there was Lakefield but even that was not this complex). I hope Ian Cutress or Anandtech does a deep dive with different laptops sometime in the future.

Slay 8 GB RAM!

 

[FlameTail]

This is not news. With current Xe graphics laptops is exactly the same. Xe Graphics with dual slot memory, UHD graphics with one slot.

Does AMD do something like this?

 

[mikk]

Here we have some die size estimations.

Based on the photos of Meteor Lake-H I did some rough estimates on the die sizes:

Compute Tile (Intel 4): 8,48 x 7,84 mm = 66,48 mm²
I/O Tile (TSMC N6): 2,96 x 9,44 mm = 27,94 mm²
SOC Tile (TSMC N6): 10,8 x 8,96 mm = 96,77 mm²
Graphics Tile (TSMC N5): 10,16 x 4,08 mm = 44,45 mm²

https://twitter.com/x/status/1736059767880085868

Twitter link doesn't work for some reason, it's from Andreas Schilling.

 

[cebri1]

Does AMD do something like this?

If they don't they should. You are starving any decent GPU with half the BW.

 

[cebri1]

Here we have some die size estimations.

https://twitter.com/x/status/1736059767880085868

GPU is tiny compared with its desktop counterpart.

 

[controlflow]

Microcenter here already shows availability of Meteor Lake laptops..8 in stock for Acer Swift Go ($999) and 3 for MSI Prestige ($1499). I was expecting a paper launch. Impressive!

Best Buy had the Zenbook 14 OLED with 155H 1 hour after launch on December 14th. Costco had the MSI 16 Prestige at the same time. The launch seems real but at the same time we don't see products from all the OEMs yet. Dell? Lenovo? HP?

 

[TESKATLIPOKA]

GPU is tiny compared with its desktop counterpart.

Yeah, but a lot of It is in I/O or SoC Tile.
If they could feed It, then 256EU IGP wouldn't be a problem either.
As I see It, with that Micron 9.6gbps LPDDR5T memory Intel could make a 192EU IGP and be able to feed It.

 

[controlflow]

Thank god. We need 8GB memory to die in 2023. Memory is not that expensive. BTW are there any laptops with LPDDR5x7466. That is the fastest supported memory right?

On Early issues even Apple iphone has had early issues which was fixed with further software updates. I think Intel was fixated on getting it out before end of the year. Probably needed more QA time for packaging team. its something unprecedented(there was Lakefield but even that was not this complex). I hope Ian Cutress or Anandtech does a deep dive with different laptops sometime in the future.

https://www.bestbuy.com/site/asus-zenbook-14-oled-14-3k-touch-laptop-intel-core-ultra-7-32gb-memory-1tb-ssd-ponder-blue/6569307.p?skuId=6569307

This one is 32GB of LPDDR5 7466 according to this review https://www.ultrabookreview.com/66513-asus-zenbook-14-ux3405-review/

 

[FlameTail]

Yeah, but a lot of It is in I/O or SoC Tile.
If they could feed It, then 256EU IGP wouldn't be a problem either.
As I see It, with that Micron 9.6gbps LPDDR5T memory Intel could make a 192EU IGP and be able to feed It.

LPDDR5T - 9600 MT/s 🔥🔥🔥

 

[FlameTail]

Here we have some die size estimations.

https://twitter.com/x/status/1736059767880085868

So total size about 240 mm².

 

[TESKATLIPOKA]

So total size about 240 mm².

Pretty big, but I don't think It's more expensive than Phoenix built on N4.

 

[TESKATLIPOKA]

Looks like I don't have a choice.

I'm going to Germany, learn German, goto Computerbase in Berlin, meet MD Frank Hüber and tell him he's wrong!

That should go well I think. 🧐

Tbh, I think they should consider testing MTL's LP E cores as well.

Writing him an email would be much faster and cheaper. Also, you don't need to learn German, I think they are pretty efficient in English. At worst just use google translate.

 

[FlameTail]

Pretty big, but I don't think It's more expensive than Phoenix built on N4.

I doubt that.

Phoenix is monolithic, whereas MTL uses advanced packaging tech, which would be expensive.

 

[TESKATLIPOKA]

I doubt that.

Phoenix is monolithic, whereas MTL uses advanced packaging tech, which would be expensive.

Would be expensive? How much of the manufacturing cost of the whole chip?

 

[eek2121]

I like DOS too. But sadly, it can be run only on VMs in UEFI systems these days.

Oh boy, let me introduce you to FreeDOS: https://freedos.org/

Difference in using faster memory is only 4%, that's good news in case Intel will later increase the EU count.

155h with 7467Mhz is 9% faster than 7840HS 6400MHz, not sure how much improvement we would see If AMD used the same memory.

My question is at what power Intel IGP is faster? 35W, 45W or more?

AMD chips show very little improvement moving from DDR5-6000 to DDR5-8800, so I would be surprised if the mobile chips saw more than single digit gains. They do have less L3 cache, however, so maybe…

I think Intel will want to keep at least 6-8 P-cores.
E-cores can increase nT performance, but you also need to keep in mind how many threads the App can load.

If you have an App which can use up to 12 threads, then 8P+4e would perform better than 4P+16E.

I can definitely see Intel moving to 6+8 or 6+16 in the future, but I have seen no leaks of core count changes as of yet outside the supposedly killed 8+32 leak, which makes sense because sharing all that bandwidth and power across 40 cores is not gonna be fun.

Microcenter here already shows availability of Meteor Lake laptops..8 in stock for Acer Swift Go ($999) and 3 for MSI Prestige ($1499). I was expecting a paper launch. Impressive!

I want a mini PC with the 185h and 32gb RAM. If someone makes it I will buy one (next year).

 

[Tigerick]

Here we have some die size estimations.

https://twitter.com/x/status/1736059767880085868

Twitter link doesn't work for some reason, it's from Andreas Schilling.

Good info, I have included the numbers in the table of frontpage. Too bad we won't know the amount of transistors which is crucial to calculate effective density of each process node....

 

[Mopetar]

Wake me up when Intel catches upto Apple in efficiency

The story of Rip Van FlameTail.

 

[SiliconFly]

So total size about 240 mm².

In a way, I think we should count the base tile as well

 

[Hulk]

I think Intel will want to keep at least 6-8 P-cores.
E-cores can increase nT performance, but you also need to keep in mind how many threads the App can load.

If you have an App which can use up to 12 threads, then 8P+4e would perform better than 4P+16E.

You are correct. Until you get to about 14 threads I think 8+4 would perform better than 4+16.

 

[H433x0n]

A bug whose correction improve perf/watt by 20%..?..
And you think that Intel engineers were unaware of the thing till launch day..?.

As far as I know, all reviewers were using a pre-release laptop. I don’t think it’s stretch that the microcode wasn’t updated to production version on a pre production product.

 

[SiliconFly]

Would be expensive? How much of the manufacturing cost of the whole chip?

I'm not very sure as I haven't come across any actual numbers, but I remember reading that MTL's Intel 16 base tile itself isn't that cheap. And foveros packaging isn't cheap either. It all adds up.

And more importantly, when they move to 20A for ARL's CPU & N3 for the GPU tile, the cost should increase significantly.

 

[mikk]

Difference in using faster memory is only 4%, that's good news in case Intel will later increase the EU count.

155h with 7467Mhz is 9% faster than 7840HS 6400MHz, not sure how much improvement we would see If AMD used the same memory.

My question is at what power Intel IGP is faster? 35W, 45W or more?

In timespy from 6400 to 7467 memory Phoenix on 780M improves 100 points, it's a 3% improvement. It depens on the game, some will gain more and some less.

He tested power efficiency in SOTTR where 155H at 28-30W pulls away slightly. There is an almost sudden drop below 25W which seems odd, maybe suboptimal power balance between CPU and GPU? He says there is a driver in the works with low power improvements expected to go live in 1 month.

 

[BorisTheBlade82]

To calculate real power efficiency, we not only need to plot the power usage curve, we also need to calculate the area under the curve and then compare ratios. No one in their right mind will waste so much of their time doing that considering the CPU power usage plots pretty much keeps fluctuating making area under the curve a nightmare to calculate. Not worth our time.

Other indirect but easier way of calculating power usage of 2 CPUs is to first ensure they have the SAME performance or at least very similar. Then running power usage tests on both CPUs designed to highlight their design goals or real world usage scenarios. Not CB.

Throwing PL1 and PL2 numbers and saying one is more power efficient than the other is just plain wrong.

That is exactly what I made this for:

Discussion - PES | Assessing Power and Performance Efficiency of x86 CPU architectures

Dear Community, so this is my first thread here as a long-time lurker - but I felt the desire to share a small hobby-project of mine from the last couple of months with you... Performance Efficiency Suite - What is it about? Most Reviewers solely focus on what they consider to be the most...

forums.anandtech.com

 

[SiliconFly]

Oh boy, let me introduce you to FreeDOS: https://freedos.org/

Sadly, I've already tried that. Even FreeDOS doesn't work with modern UEFI. Requires legacy BIOS thats not available anymore. I think Intel declared Legacy BIOS as dead many years ago.