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.)

 

[tamz_msc]

https://twitter.com/x/status/1758875038130794977

What on earth is IDC doing?

There seems to be a typo in there:

Lion Cove:

64 KB L1i + 48 KB L1D = 112 KB * 4 = 448 KB L1

Skymont:

64 KB L1i + 32 KB L1D = 96 KB * 4 = 384 KB L1

Total = 448 + 384 = 832 KB L1

 

[adroc_thurston]

But maybe that's why AMDs market share is so low in the notebook market

Que? It went up YoY and QoQ.

What on earth is IDC doing?

Adding more stuff?

 

[Abwx]

No question that AMD has a marginal lead on compute and graphics efficiency over Meteor Lake. Unfortunately for AMD, that doesn't count for much. There being a laptop available in the right size, form factor, and screen for the right price is what matters to the overwhelming majority of consumers, not whether the sticker says Intel or AMD. I point that out because while it's fun to discuss the merits of one chip versus another there shouldn't be any expectation of such being reflected in the market.

Marginal..?..
Now that we have numbers that are no more blured with excessive turbos we can see at what point it s "marginal", in CB R15 at 24W the 155H score 1400pts while the 7840U score 1800pts, and for good measure a 6800U score 1640pts@25W, actually a 6 + 8 MTL doesnt even manage to match a Rembrandt in CPU perf/watt, that tell it all about Intel 4 efficency.

And using TSMC for their GPU and 33% more SPs doesnt even save their days,
they are still behind in GPU perf and perf/watt, as usual their saving grace was huge turboing.

At some point Intel will no more benefit from their grip on OEMs, guess that it takes a lot of shady efforts and prices reductions to get them on board at the current rates, because AMD chips are massively available in Handhelds, so it s a certainity that there s something else than availability in the laptop market.

https://www.notebookcheck.net/HP-Spectre-x360-14-review-High-end-convertible-now-with-a-larger-120-Hz-OLED.804502.0.html

https://www.notebookcheck.net/Lenovo-ThinkPad-T14s-G4-review-Business-laptop-is-better-with-AMD-Zen4.763581.0.html

https://www.notebookcheck.net/Lenovo-Yoga-7-14-G7-review-Multimedia-convertible-shines-with-AMD-Ryzen-7-6800U.645586.0.html

 

[adroc_thurston]

At some point Intel will no more benefit from their grip on OEMs, guess that it takes a lot of shady efforts and prices reductions to get them on board at the current rates, because AMD chips are massively available in Handhelds, so it s a certainity that there s something else than availability in the laptop market.

Nothing shady there, Intel does what Intel does best.

 

[Saylick]

Nothing shady there, Intel does what Intel does best.

Like overly optimize for certain industry-wide accepted benchmarks?

Industry group invalidates 2,600 official Intel CPU benchmarks — SPEC says the company's compiler used unfair optimizations to boost performance

It mostly affects Sapphire Rapids benchmarks.

www.tomshardware.com

 

[H433x0n]

I’m convinced that laptop performance & efficiency benchmarks are completely useless. It’s like choose your own adventure bad faith edition. There’s so many variables that CPU performance can vary by like 30-40%.

There’s people in this thread saying MTL gets better battery life (normalized to battery capacity) than Phoenix and another saying MTL is less efficient than Rembrandt both with supporting data

 

[igor_kavinski]

Intel® Core™ Ultra Mobile Processors (Series 1) Product Brief

Intel® Core™ Ultra Mobile Processors (Series 1) deliver an optimal balance of performance and power efficiency, featuring 3D performance hybrid architecture and advanced AI capabilities.

www.intel.com

OK, I will give Intel the benefit of the doubt for the time being coz there are no 165U laptops out yet.

But still,



While for 7840U:



Intel setting an upper power limit of 57W says a lot, no?

 

[Markfw]

I’m convinced that laptop performance & efficiency benchmarks are completely useless. It’s like choose your own adventure bad faith edition. There’s so many variables that CPU performance can vary by like 30-40%.

There’s people in this thread saying MTL gets better battery life (normalized to battery capacity) than Phoenix and another saying MTL is less efficient than Rembrandt both with supporting data

I will say one thing... I am so confused by the benchmarks, I am calling them equal or AMD gets a little nod. This is good for Intel, as in all other areas, they are in the dumper. desktop, server and HEDT are clearly won by AMD in every respect, performance, efficient, etc....

 

[H433x0n]

I will say one thing... I am so confused by the benchmarks, I am calling them equal or AMD gets a little nod. This is good for Intel, as in all other areas, they are in the dumper. desktop, server and HEDT are clearly won by AMD in every respect, performance, efficient, etc....

I don’t doubt Phoenix outperforms it below 25-30W but even now 2 months post launch I have no idea what the actual performance delta is.

The only thing I’m sure of is that once every few weeks when NotebookCheck reviews a poorly performing MTL laptop the review will be posted here for partisans to spike the football over.

I’m going to be bold and assume that when LNL & ARL mobile launches the same thing will happen then too.

 

[H433x0n]

Intel® Core™ Ultra Mobile Processors (Series 1) Product Brief

Intel® Core™ Ultra Mobile Processors (Series 1) deliver an optimal balance of performance and power efficiency, featuring 3D performance hybrid architecture and advanced AI capabilities.

www.intel.com

OK, I will give Intel the benefit of the doubt for the time being coz there are no 165U laptops out yet.

But still,

View attachment 94122

While for 7840U:

View attachment 94123

Intel setting an upper power limit of 57W says a lot, no?

I don’t think so. What’s the harm in allowing bursts for 3-5 seconds of up to 57W while not on battery? If the chassis can support it and battery life isn’t an issue, you’d want that capability for enhanced performance.

 

[poke01]

I don’t think so. What’s the harm in allowing bursts for 3-5 seconds of up to 57W while not on battery? If the chassis can support it and battery life isn’t an issue, you’d want that capability for enhanced performance.

That’s the issue. Upwards of 57 watts, no longer makes this CPU an efficiency focused CPU. Why does need to go up 57 watts anyway even for a few seconds?

It just shows that Intel can’t do low power CPUs. Hopefully Lunar changes this. This boosting behaviour is just not cut out for thin and light laptops which is what the U series processors are meant for.

 

[Thibsie]

I don’t think so. What’s the harm in allowing bursts for 3-5 seconds of up to 57W while not on battery? If the chassis can support it and battery life isn’t an issue, you’d want that capability for enhanced performance.

Generally speaking, on can be discussed.
For a U CPU supposed to be optimized for efficiency? LOL

 

[igor_kavinski]

While we wait for Intel to get their act together, Hawk Point got more power efficient than 7840U: https://videocardz.com/newz/amd-ryz...-phoenix-and-steam-decks-van-gogh-at-10-watts

 

[coercitiv]

Intel setting an upper power limit of 57W says a lot, no?

Turbo power and cTDP are not the same thing. You should keep in mind there are AMD laptops out there that mimic Intel's behavior, mostly because OEMs nowadays have multiple power modes on their machines and "Performance" is used to score well in reviews while normal people use Balanced or Cool etc.

Spoiler: boost vs. sustained

What’s the harm in allowing bursts for 3-5 seconds of up to 57W while not on battery?

It ruins the user experience in terms of noise and heat because we're talking about a ~100% increase in power draw relative to the TDP target of a cooling system that is already designed to run fans on high at nominal power. To put things into perspective, there was a time when ~57W was the boost for 47W TDP CPUs. Intel had a good rule of thumb of pushing boost power around 20-30% higher than TDP target, and it worked well, extracting more performance for a modest increase in noise.

I can see a 100% power boost ratio being ok in devices with passive cooling or active cooling with very low noise at nominal power, in these cases the boost would likely tap solely on available thermal capacity, assuming skin temps are also taken into account.

 

[tamz_msc]

It ruins the user experience in terms of noise and heat

It really doesn't. I use mine plugged in all the time.

Besides, it also consumes less energy for the 2-3 second task at hand.

 

[DavidC1]

There seems to be a typo in there:

Lion Cove:

64 KB L1i + 48 KB L1D = 112 KB * 4 = 448 KB L1

Skymont:

64 KB L1i + 32 KB L1D = 96 KB * 4 = 384 KB L1

Total = 448 + 384 = 832 KB L1

Does Windows actually detect or report what the CPU has or read data from what programmers wrote in? Because in the latter case it's just people entering information based on what they know at the time, then Bionic_Squash would be right. Also, he's among the reliable leakers. 64KB L1i for P cores is just what's in Redwood Cove.

Also, it's not showing the 8MB system level cache either.

GPU-Z does the same thing. The specs that are "shown" by GPU-Z is people working for TPU manually entering data in. That's why GPU-Z was consistently giving out wrong info for Intel iGPUs up until very recently.

 

[TESKATLIPOKA]

It really doesn't. I use mine plugged in all the time.

Besides, it also consumes less energy for the 2-3 second task at hand.

It all depends on what notebook you have.

I have a 14-inch laptop with PHX and thank god It has a 2 fan cooling system. It can cool even 60W sustained without being too noisy.
If I bought Asus Zen laptop with PHX with only a single fan, then I would be worse off.

The same applies to Intel laptops, but I must say that >100% increase in TDP during boost is not very appealing even If only for a few seconds.

 

[DavidC1]

It really doesn't. I use mine plugged in all the time.

Besides, it also consumes less energy for the 2-3 second task at hand.

I used a Dell XPS 12 "Ivy Bridge" ultrabook for few years.

It didn't take a lot for the CPU to reach it's max 2.8GHz Turbo. Enough web browsing tabs would do it. Also, so many threads and background applications don't throttle down lot of the time, so it's consuming resources, adding to the problem.

I didn't have such an issue with my 2600K desktop system. I guess the desktop system is so much faster so it doesn't need to Turbo most of the time. You'd think a Turbo on the laptop does the same thing but you have slower SSD, PCH, graphics, CPU, plus all layers of power management which in theory should save power, and in overall it is, but not as much as we thought.

While in theory the "HUGI" concept is great, on Windows it only works about half the time. Oh how many times I have searched for why this particular windows process is eating up 30% of my CPU. Or how on laptops when you first get it, it takes as long as an hour before all updates are finished and the CPU settles down to the low power state, wondering the whole time why your brand spanking new laptop is using so much battery?

First thing I see in a laptop review is to check if it has a 1080p screen, because all other options waste power. On my current device I run Youtube at max 720p, or sometimes 480p, because it cuts down CPU power by 3/4W, even though it has VP9 acceleration. Then I install Throttlestop, and then fire up services.msc to prevent lots of processes from starting. Then taskbar to reduce reduce clutter such as Intel graphics tray, OneDrive, Bluetooth, etc. How many average users will do that?

Which is why the concept of a dedicated LPE core on Meteorlake sounded great. Except it doesn't seem to be so good at relocating those pesky real-world tasks, meaning it has to fire up the power hungry tCPU tile anyway.

 

[coercitiv]

It really doesn't. I use mine plugged in all the time.

How high does your TGL boost in that Inspiron 14? On the one review I was able to find on notebookcheck (5140 2-in-1 with TGL UP3), CPU package power was under 40W even during boost, and the entire system did not pull more than 55W peak. Seems to me that Dell is using the more conservative 30% boost I was talking about.

 

[DavidC1]

How high does your TGL boost in that Inspiron 14? On the one review I was able to find on notebookcheck (5140 2-in-1 with TGL UP3), CPU package power was under 40W even during boost, and the entire system did not pull more than 55W peak. Seems to me that Dell is using the more conservative 30% boost I was talking about.

Concept of spec sheet on laptop CPUs are irrelevant. XPS 12 3517U had a Turbo of 2.8GHz, not 3.0.

It may be called 15W or 28W but laptop makers are given freedom to adjust TDP levels to almost any degree in between and above. From a design concept it's great, but it nullifies lots of the points CPU manufacturers make in their marketing presentations.

My Yoga's Kaby is set at 7W PL1 and 15W PL2 sometimes, but other times I've seen it go low as 5W PL1 and 7W PL2. It varies on tremendous other factors such as what you are doing, whether it's on clamshell or tablet mode. Based on recent reviews, if anything the variation has increased. PL1/PL2 specs are meaningless because it's a dynamic figure. Sometimes it's 28/28, other times it's 28/45, it can also be 15/28 all with the same device.

 

[tamz_msc]

Does Windows actually detect or report what the CPU has or read data from what programmers wrote in? Because in the latter case it's just people entering information based on what they know at the time, then Bionic_Squash would be right. Also, he's among the reliable leakers. 64KB L1i for P cores is just what's in Redwood Cove.

Also, it's not showing the 8MB system level cache either.

GPU-Z does the same thing. The specs that are "shown" by GPU-Z is people working for TPU manually entering data in. That's why GPU-Z was consistently giving out wrong info for Intel iGPUs up until very recently.

Does WIndows even detect SLC at all?

One way to find out would be to check what Windows on arm devices report, like the 8cx or the new, unreleased Snapdragons.

 

[DavidC1]

Does WIndows even detect SLC at all?

One way to find out would be to check what Windows on arm devices report, like the 8cx or the new, unreleased Snapdragons.

Exactly is my point.

A detection of an actual resource within the CPU would require either of the two things: The firmware within the CPU tells the OS what it has, and the OS adds extra line to recognize that and report it. Second, by running a microbenchmark, and try to see what's there. I doubt the latter is done, so it's the former. Plus benchmarks take time, and reporting is instant.

192KB L1D would be another thing that Lunarlake is "copying" Apple.

It seems the #1 trick to power efficient and powerful CPUs are caches, and caches everywhere. uOP caches, L3 caches, more registers, memory side caches, decoder queues and BTBs.

Someone should modify the meme "moar cores" to "moar cache".

 

[tamz_msc]

How high does your TGL boost in that Inspiron 14? On the one review I was able to find on notebookcheck (5140 2-in-1 with TGL UP3), CPU package power was under 40W even during boost, and the entire system did not pull more than 55W peak. Seems to me that Dell is using the more conservative 30% boost I was talking about.

I can just test it at whatever PL1/PL2 combination I want using ThrottleStop:

I choose WebXPRT 4, because it is mostly single threaded and also runs for a long time compared to other benchmarks.

30W PL2/25 W PL1:



Package power was 29.4 W maximum, with max package temperature 90 degrees Celsius.

51 W PL2/ 25 W PL1:



Max package power was 41.4 W, and max package temperature was 91 degrees Celsius.

Most of the time was spent running at 4-7 W during the test in both configurations.

 

[igor_kavinski]

You should keep in mind there are AMD laptops out there that mimic Intel's behavior

Would be nice to see a review of such a laptop.

 

[Abwx]

Would be nice to see a review of such a laptop.

There s no such laptops, either that s 45W HS variants that can boost comfortably to 60W or even more or that s U series wich are limited to 30W or so.

There s no mainstream laptop with a U SKU that boost to more than 30W if we except the special cases of the Framework or the Ayaneo Kun Handeld, and still the Framework laptop use 50-55W at most for the full device, AMD has segmented the TDPs such that the U dont reach H variants TDPs levels.

https://www.notebookcheck.net/AMD-Ryzen-7-7840U-Processor-Benchmarks-and-Specs.716412.0.html