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.)

 

[H433x0n]

If this chart was comparable to reality, then MTL at 3.3GHz would consume ~40% of Raptor. So If Raptor needed 45W MTL would need ~16W.

It’s a bit more complicated since the graph referenced is likely of a single core whereas it takes more energy to “spool up” a 6+8 die and the uncore components.

 

[Hitman928]

These laptops being compared have different battery sizes and chassis. We need an apples/apples test with both of these chips in the same chassis like what JarrodTech did with 13700H vs 7840HS.

The only improvement we should expect is what is shown below.

View attachment 90068

Lots of the uncore/IO of the chip won't see this kind of improvement so you won't get this straight forward of benefit at a chip level. Additionally, it seems that Intel has eaten away any efficiency gains from the process with the complexity overhead of the tile design.

 

[adroc_thurston]

but from a marketing perspective, something is better than nothing so comparing to RPL

I'm pretty sure they actually marketed RPL + Keem Bay m.2 addon for a bit.

 

[Hitman928]

I'm pretty sure they actually marketed RPL + Keem Bay m.2 addon for a bit.

Never saw anything like that on the consumer side, must of just been to system builders/integrators. I'm guessing no one bit since there haven't been any end products with that configuration, at least that I'm aware of.

 

[SiliconFly]

If MTL has worse or ~ same perf/watt than RPL, it's gonna be ICL all over again lol

There is one misconception about MTL that keeps popping up pretty often. MTL's PPW under full load will definitely be closer to RPL. Absolutely no doubts there. It's the same exact RPL core in a fancy new tile.

MTL's efficiency is not about when the CPU tile runs under full load. It's all about when the CPU tile is NOT running under full load. Especially when the CPU & GPU tiles are off which is most of the time for most users.

Also. MTL is guaranteed to fall flat in power-efficiency benchmarks that stresses the CPU tile cos MTL isn't designed for that. We'll be needing new benchmarks. Either way, real world data will be available in a few days from now.

Imho, MTL's efficiency is all about lighter real-world loads only. Nothing else.

 

[adroc_thurston]

Imho, MTL's efficiency is all about lighter real-world loads only. Nothing else.

wow nice goalpost move.
It sucks at that too since SA power stinks and will stink until LNL-ish.

Never saw anything like that on the consumer side

Yeah it was commercial OEM stuff only.

 

[Hitman928]

Especially when the CPU & GPU tiles are off which is most of the time for most users.

I'm still skeptical of this claim. Anything interactive, apps, web browsing, emails, are going to be run on a P-core for responsiveness. I don't think most users will see much benefit from the new ELP cores. Maybe in long idle battery or video playback scenarios but even then, I'm not sure how much better it may be versus a monolithic design that can power down pretty much anything it needs to anyway. Obviously we'll have to wait for the tests to see, but I'm not that hopeful on this point.

 

[H433x0n]

I'm still skeptical of this claim. Anything interactive, apps, web browsing, emails, are going to be run on a P-core for responsiveness. I don't think most users will see much benefit from the new ELP cores. Maybe in long idle battery or video playback scenarios but even then, I'm not sure how much better it may be versus a monolithic design that can power down pretty much anything it needs to anyway. Obviously we'll have to wait for the tests to see, but I'm not that hopeful on this point.

That’s no longer the case with MTL. They’ve changed to taking an Apple approach by changing how quickly it ramps clocks and what cores are targeted for certain tasks. It doesn’t automatically target P cores first anymore. It begins with LP e-cores -> e-cores -> p-cores.

 

[SiliconFly]

wow nice goalpost move.

No goalpost moves. It's hard to understand MTL until you actually look into it's design goals.

Like I said, it's primary purpose is to improve efficiency using the LP E cores in lighter real-world loads and nothing else. Thats where the 2X mythical efficiency claims come from.

"MTL's efficiency is not about when the CPU tile runs under full load. It's all about when the CPU tile is NOT running under full load. Especially when the CPU & GPU tiles are off which is most of the time for most users."

Stress testing the CPU tile is just plain wrong and useless in case of MTL as it doesn't work that way. It will fail.

Did it ever occur to you that since the LP E cores will be active most of the time, the most important power curve we should be considering is that of the LP E cores and not the MLID's useless power curve of the CPU tile itself? MLID is wrong which is understandable, but you are better than that.

 

[Hitman928]

That’s no longer the case with MTL. They’ve changed to taking an Apple approach by changing how quickly it ramps clocks and what cores are targeted for certain tasks. It doesn’t automatically target P cores first anymore. It begins with LP e-cores -> e-cores -> p-cores.

That will be interesting to follow. I believe it was AMD's 4000 series that tried a slower ramp with the cores to save power and they got harshly criticized for it. If they're going through multiple layers to then ramp the p-cores, it may not be well received. Maybe they'll handle it better though, we'll see. Even still, I imagine the vast majority of what users do on a computer will quickly shuffle to the P-core, so I still don't know if it will be a saving grace here but if Intel pulls it off well, I'm all for it.

 

[Hitman928]

No goalpost moves. It's hard to understand MTL until you actually look into it's design goals.

Like I said, it's primary purpose is to improve efficiency using the LP E cores in lighter real-world loads and nothing else. Thats where the 2X mythical efficiency claims come from.

"MTL's efficiency is not about when the CPU tile runs under full load. It's all about when the CPU tile is NOT running under full load. Especially when the CPU & GPU tiles are off which is most of the time for most users."

Stress testing the CPU tile is just plain wrong and useless in case of MTL as it doesn't work that way.

Did it ever occur to you that since the LP E cores will be active most of the time, the most important power curve we should be considering is that of the LP E cores and not the MLID's useless power curve of the CPU tile itself? MLID is wrong which is understandable, but you are better than that.

The battery tests shown in the leak suggest either this configuration isn't working correctly (possible especially if it was under Win10) or it's not having the desired effect. Looking forward to actual trusted 3rd party reviews.

 

[adroc_thurston]

It's hard to understand MTL until you actually look into it's design goals.

Yes it's a schizophrenic product made by morons way too into advanced packaging.
LNL backtracks on all that mess and does the usual SOC + PCH scheme.

or it's not having the desired effect.

Well yeah, 2 LITTLEs isn't enough anyway.

 

[Geddagod]

MTL's PPW under full load will definitely be closer to RPL. Absolutely no doubts there. It's the same exact RPL core in a fancy new tile.

It's on a new node. Free PPW.

Also. MTL is guaranteed to fall flat in power-efficiency benchmarks that stresses the CPU tile cos MTL isn't designed for that

It literally is. The CPU tile is on a new node.

 

[SiliconFly]

The battery tests shown in the leak suggest either this configuration isn't working correctly (possible especially if it was under Win10) or it's not having the desired effect. Looking forward to actual trusted 3rd party reviews.

Those HXL leaks are stress tests where MTL will surely fail, cos MTL's efficiency drops like a rock in these type of stress tests. Other way of looking at it is, if in a efficiency benchmark, MTL's P cores are active most of the time, MTL will surely fail that test and will perform similar or worse than RPL.

A new benchmark that prioritizes MTL's LP E cores will actually give better results. But one such benchmark doesn't exist yet. Until then, we'll have to rely on real world numbers when MTL releases in a few days.

 

[SiliconFly]

Yes it's a schizophrenic product made by morons way too into advanced packaging.

Now we know a little about you
You need to stop with the personal attacks. This is not acceptable behavior. -AT Moderator Shmee

 

[Geddagod]

Those HXL leaks are stress tests where MTL will surely fail, cos MTL's efficiency drops like a rock in these type of stress tests

Which is embarrassing

Other way of looking at it is, if in a efficiency benchmark, MTL's P cores are active most of the time, MTL will surely fail that test and will perform similar or worse than RPL.

It's on a better node, if it does so it's literally embarrassing lol

 

[SiliconFly]

It's on a new node. Free PPW.

It literally is. The CPU tile is on a new node.

It's on a better node, if it does so it's literally embarrassing lol

There will be some node gains. But 2X efficiency gains is just not possible with node gains alone. If they ever hit 2X, it'll be due to the new architecture & the LP E cores only.

 

[Geddagod]

There will be node gains. But 2X is not possible with node gains alone. If they ever hit 2X, it'll be due to the new architecture & the LP E cores only.

One would hope there should be at least >10% gains from the node. You literally just said MTL will perform similarly or worse than RPL iso power.

 

[Hitman928]

Those HXL leaks are stress tests where MTL will surely fail, cos MTL's efficiency drops like a rock in these type of stress tests. Other way of looking at it is, if in a efficiency benchmark, MTL's P cores are active most of the time, MTL will surely fail that test and will perform similar or worse than RPL.

A new benchmark that prioritizes MTL's LP E cores will actually give better results. But one such benchmark doesn't exist yet. Until then, we'll have to rely on real world numbers when MTL releases in a few days.

I'm talking about the Bilibili leak, not the HXL leak. The Bilibili leak had actual battery life tests, not stress tests, and MTL did not have an impressive showing. That said, there are a lot of details left out but it's not a great first impression.

 

[Hitman928]

@Geddagod hope your studying went well. Good luck on your finals. Very grateful to be done with that part of my life, lol.

 

[SiliconFly]

One would hope there should be at least >10% gains from the node. You literally just said MTL will perform similarly or worse than RPL iso power.

RPL is Intel 7++

Efficiency gains from Intel 7++ to Intel 4 should be very less.

 

[Geddagod]

@Geddagod hope your studying went well. Good luck on your finals. Very grateful to be done with that part of my life, lol.

Thx, got nothing too hard this semester, cuz I'm just finishing up a bunch of general education requirements lol.
Next semester tho I'm taking a digital logic course, which I heard requires 20-30 hours a week

 

[H433x0n]

One would hope there should be at least >10% gains from the node. You literally just said MTL will perform similarly or worse than RPL iso power.

There probably will be, we don't have any real data yet. We're comparing across different laptops with different power delivery and battery capacities. We're also using data from Core 7 155H that’s a 28W processor that's 3 tiers below the top SKU.

We don’t even have perf/watt graphs. Just screenshots of benchmark scores and rumors of how much power was used.

I'm talking about the Bilibili leak, not the HXL leak. The Bilibili leak had actual battery life tests, not stress tests, and MTL did not have an impressive showing. That said, there are a lot of details left out but it's not a great first impression.

The battery life test compared laptops with different battery capacities.

 

[H433x0n]

RPL is Intel 7++

Efficiency gains from Intel 7++ to Intel 4 should be very less.

The node itself is fine. There’s really no need to jump to conclusions yet. We need to compare same laptop with both RPL & MTL under same conditions to get real data.

 

[cebri1]

The node itself is fine. There’s really no need to jump to conclusions yet. We need to compare same laptop with both RPL & MTL under same conditions to get real data.

Also, some of these test apparently were done on Win10 and I wouldn't be surprised if the new thread director was not used either.