Intel Meteor Lake Technical Deep Dive

[DavidC1]

Right I forgot all about Broadwell-Y, I think I only remembered it as Core M.

Sorry for the bit of rant, but 14nm was terrible. And it was delayed 6 months. Only reason most people don't seem to remember is because 10nm problems dwarfed it.

After all the hype Core M was a disappointment.

Intel’s Core M Strategy: CPU Specifications for 9mm Fanless Tablets and 2-in-1 Devices

www.anandtech.com

Intel Broadwell Architecture Preview: A Glimpse into Core M

www.anandtech.com

Look at this damned slide.

https://images.anandtech.com/doci/8355/BDW-14nm.png

Then this which got everyone excited: https://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain

In hindsight, the above "benchmark" is akin to gameplay demos CD Projekt RED did with the press and "gameplay". In reality the chip could only achieve those levels for about 30 seconds and performance would drop to 2/3rds.

Rule of Thumb: The more they talk about it and hype it the more it'll suck. Cyberpunk 2077, anyone remember that? I've never seen in the history of gaming or any other product something that has been hyped for so long and with such vigor. Of course I would say it crashed as hard as it was hyped too.

 

[las]

Do you remember the Arrowlake performance leak a little while ago? It was disappointing.

Apparently not only it'll have N3 and 20A versions, but Redwood Cove + Crestmont version and Lion Cove + Skymont version. The ISA documents also support this with -S and non-S models supporting different instructions. Credible leakers pointed to -P mobile having 20A and -S being on N3. Incredibly, some even say low power "Arrowlake" is potentially Meteorlake but on Intel 3. But that won't really matter as Lunarlake is the Real Thing for low power anyway.

No ADM in Meteorlake. Kepler_L2 has said ALL ADM models are cancelled. Suffice to say extra cache module layer would be a waste of money on 8 Xe core part since AMD is also doing fine without it.

The only thing that seems noticeable for P is the doubled L1i cache to 64KB. It's definitely a departure from the server P core in Granite Rapids: https://images.anandtech.com/doci/20034/HotChips 2023 Press Briefing Final__07.png

The E cores are different too.
Crestmont client: https://images.anandtech.com/doci/20046/Architecting Our Next Gen Power Efficient Processor_FINAL CLEAN-13.png

Sierra Glen server: https://images.anandtech.com/doci/20034/HotChips 2023 Press Briefing Final__08.png

Most notable, Crestmont gets Allocate/Rename/Move Elimination/Zero Idiom increased to 6-wide while Sierra Glen stays at 5. Chipsandcheese overview shows that at longer loop length, the decode throughput is limited by the renamer: https://chipsandcheese.com/2021/12/21/gracemont-revenge-of-the-atom-cores/

So Crestmont in Meteorlake should get full 6-wide throughput. Slides also mention branch prediction on Crestmont but not on Sierra Glen. Apparently Sierra Glen prioritizes higher clocks and Crestmont, higher perf/clock.

Yes I do and it delivered +20% MT while (probably) using an ES sample with lower clocks than the final products.

I don't trust leaks anyway. I don't think Intel will use TSMC 3nm for CPU but GPUs, we will see tho

 

[Dr. Dro]

The next stop for me is likely the Lunar Lake platform or its desktop equivalent. Should be 17th Gen chips, I suppose.

 

[DavidC1]

Yes I do and it delivered +20% MT while (probably) using an ES sample with lower clocks than the final products.

I don't trust leaks anyway. I don't think Intel will use TSMC 3nm for CPU but GPUs, we will see tho

That slide didn't say anything about being an ES slide. No company does official benchmarking on an ES.

 

[Squared]

Sorry for the bit of rant, but 14nm was terrible. And it was delayed 6 months. Only reason most people don't seem to remember is because 10nm problems dwarfed it.

After all the hype Core M was a disappointment.

Intel’s Core M Strategy: CPU Specifications for 9mm Fanless Tablets and 2-in-1 Devices

www.anandtech.com

Intel Broadwell Architecture Preview: A Glimpse into Core M

www.anandtech.com

Look at this damned slide.

https://images.anandtech.com/doci/8355/BDW-14nm.png

Then this which got everyone excited: https://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain

In hindsight, the above "benchmark" is akin to gameplay demos CD Projekt RED did with the press and "gameplay". In reality the chip could only achieve those levels for about 30 seconds and performance would drop to 2/3rds.

Rule of Thumb: The more they talk about it and hype it the more it'll suck. Cyberpunk 2077, anyone remember that? I've never seen in the history of gaming or any other product something that has been hyped for so long and with such vigor. Of course I would say it crashed as hard as it was hyped too.

Core M allowed fanless tablets and laptops that ran Windows and outperformed Atom processors by a good margin. It would allow a tablet to load web pages with the speed of an i5 processor. It was tiny. Sure it couldn't sustain i5 performance for more than a few seconds, but it easily outshined all other options for its intended devices. I suspect the reason Intel doesn't have Core M today is because it runs Windows which isn't a great tablet OS, and it cost a lot of money to design specially, so it had too small of a market considering its research and development budget.

Intel 14nm was late by 6 months or a year, but it still came to market before the competition. Compare that to Intel 4, which is already that far behind and is still months away from release.

 

[DavidC1]

Core M allowed fanless tablets and laptops that ran Windows and outperformed Atom processors by a good margin. It would allow a tablet to load web pages with the speed of an i5 processor. It was tiny. Sure it couldn't sustain i5 performance for more than a few seconds, but it easily outshined all other options for its intended devices. I suspect the reason Intel doesn't have Core M today is because it runs Windows which isn't a great tablet OS, and it cost a lot of money to design specially, so it had too small of a market considering its research and development budget.

It doesn't matter. The comparison is the previous generation Core chips. You could get fanless using SDP on Ivy Bridge, and you had Haswell with the battery life improvements. They were talking like Core M was some revolutionary thing, when it was not.

Intel Core M-5Y70 "Broadwell" Review

Comprehensive review of Intel's new Broadwell generation with numerous benchmarks and analyses of the Core M-5Y70.

www.notebookcheck.net

-Little bit better CPU performance compared to Haswell Y
-Barely faster graphics than Haswell Y
-About equal battery life

For laptops that cost $1000 it was a PoS but all their Y lines were. It wasn't until Skylake-M there were some glimmer of hope.

The Atom actually had very decent battery life comparable to ARM. And when they moved to OoOE with Silvermont and 22nm, performance was usable. My Dell Venue 8 Pro idled at 1.2W with screen on and I got 6-8 hours battery life with a puny 17WHr battery.

I was Intel to stop BSing and actually deliver on battery life with Meteorlake. All the paper specs are worthless if the laptops don't get a drastic increase in battery life. It needs to improve significantly just to catch up to AMD, nevermind ARM.

 

[Borc]

For serious fanless sub 10W designs we have to wait for Lunar Lake, Intel promises a breakthrough in perf/w. As for Meteor Lake it should have a much better efficiency/battery life over Alder Lake and Raptor Lake. Not only Intel 4 helps, also the separate voltage rails (finally) between P and E cores is a big help. In low load scenarios they can shut down the compute tile.

 

[DavidC1]

As for Meteor Lake it should have a much better efficiency/battery life over Alder Lake and Raptor Lake.

It better, because AMD is way ahead of them now, and Meteorlake needs a significant boost just to be competitive.

In theory Meteorlake has everything to be way better than AMD's Phoenix, but it all depends on execution. From what I heard over the years Meteorlake is "meh".

Lakefield had all the basics(I mean the high level specs) that sounded like the battery life would be great, but it did no better than on-package PCH Y chips. Also they started degrading since Tigerlake. Tiger/Alder/Raptor at first glance had no reason to be significantly worse successively but in real world they are.

Not only Intel 4 helps, also the separate voltage rails (finally) between P and E cores is a big help. In low load scenarios they can shut down the compute tile.

The bottleneck in Intel platforms aren't voltage rails or process, but the PCH not being on-die. Which is why I had hoped for Lakefield to do great, but it hasn't, not even a little bit despite the PCH being connected by Foveros.

Intel 4 will only help in performance, not battery life, which is dominated by the ability to go low power when doing nothing and staying at the lowest levels as necessary. Icelake was super great at reaching really low idle, but it got kicked out of it's low power state so fast that in the real world there were zero gains for battery life. At least it did not do worse like Tiger/Alder/Raptor.

So the PPT presentations are nice to look at but it won't be until about Jan-Feb when real laptops are available for reviews and for people to buy whether Meteorlake's power management slides are worth a crap or not. For Lunarlake and other future products I have a higher hope but Meteorlake was developed during a time when there was chaos in Intel's management with lots of cutbacks and delays.

 

[LuxZg]

In short, agree with David. Looks great on paper (even though I prefer AMD) and I honestly hope it gets executed well. But only once we get access to real hardware will we see if it's just on paper or was it executed as well as it was planned. It has all the necessary for being good CPU. We'll see I guess.

 

[Borc]

It better, because AMD is way ahead of them now, and Meteorlake needs a significant boost just to be competitive.

In theory Meteorlake has everything to be way better than AMD's Phoenix, but it all depends on execution. From what I heard over the years Meteorlake is "meh".

Lakefield had all the basics(I mean the high level specs) that sounded like the battery life would be great, but it did no better than on-package PCH Y chips. Also they started degrading since Tigerlake. Tiger/Alder/Raptor at first glance had no reason to be significantly worse successively but in real world they are.

The bottleneck in Intel platforms aren't voltage rails or process, but the PCH not being on-die. Which is why I had hoped for Lakefield to do great, but it hasn't, not even a little bit despite the PCH being connected by Foveros.

Intel 4 will only help in performance, not battery life, which is dominated by the ability to go low power when doing nothing and staying at the lowest levels as necessary. Icelake was super great at reaching really low idle, but it got kicked out of it's low power state so fast that in the real world there were zero gains for battery life. At least it did not do worse like Tiger/Alder/Raptor.

So the PPT presentations are nice to look at but it won't be until about Jan-Feb when real laptops are available for reviews and for people to buy whether Meteorlake's power management slides are worth a crap or not. For Lunarlake and other future products I have a higher hope but Meteorlake was developed during a time when there was chaos in Intel's management with lots of cutbacks and delays.

Phoenix is a little dissapointing in this regards because battery life went down compared to Rembrandt. Meteor Lake is mainly about power efficiency (and graphics+NPU), there are so many improvements targeted for improved efficiency and battery life, they say the soc power is cut in half compared to Raptor Lake. I don't think it will be revolutionary when it comes to battery life but AMD isn't either.

I believe Lunar Lake will be a different beast but this chip is optimized for ultra low power from the beginning and only features 4+4 cores, it won't scale up to 45W like Meteor Lake.

Lakefield couldn't succeed with just 1 big core and subpar Tremont cores on the subpar 10nm from Icelake, with this low performance it never was appealing for OEMs and a broader mass of people no matter how good the battery life would have been. Maybe they learned from Lakefield which helped Meteor Lake though.

 

[DavidC1]

Lakefield couldn't succeed with just 1 big core and subpar Tremont cores on the subpar 10nm from Icelake, with this low performance it never was appealing for OEMs and a broader mass of people no matter how good the battery life would have been. Maybe they learned from Lakefield which helped Meteor Lake though.

Tremont was very good. It was the Lakefield implementation that sucked. In fact it was so bad, that the Tremont-only Jasperlake beat the big Sunny Cove cores in single thread by 20-30%.

Processor Comparison - Head 2 Head

Intel Pentium Silver N6005 vs Intel Core i5-L16G7 - Benchmarks, Tests and Comparisons

www.notebookcheck.net

Since Jasper lake devices only cost $300-400, you'd ignore Lakefield completely. Faster single thread, way faster multi-thread and graphics and 1/3rds the cost.

If at least it had battery life greatly improved like the original Tablet Atoms, yea it would have had a niche. But Lakefield had no place anywhere. This is what I mean. Intel can talk all the want and flash PPT presentations in your face until you get a seizure. Well, how does it perform in reality?

Meteor Lake is mainly about power efficiency (and graphics+NPU), there are so many improvements targeted for improved efficiency and battery life, they say the soc power is cut in half compared to Raptor Lake. I don't think it will be revolutionary when it comes to battery life but AMD isn't either.

See that's the thing. Meteorlake has all the potential to have a revolutionary battery life on par with ARM platform. TDP barely matters when it comes to most battery life scenarios. Web browsing, watching Youtube, peak power just doesn't matter. The ability to go to ultra low idle as a platform(not just the SoC) and be able to keep it there most of the time in real world scenarios is what gets great battery life.

I believe Lunar Lake will be a different beast but this chip is optimized for ultra low power from the beginning and only features 4+4 cores, it won't scale up to 45W like Meteor Lake.

TDP don't matter and neither does 4+4 or 8+8 or n+a. That mattered 25 years ago, before advanced power management techniques such as on-die PM chips and power gating of unused blocks. Sure, if you are gaming or rendering but no one measures battery life doing that. If you really cared in those scenarios you'd just get a 7W Y-class chip.

Battery life in laptops mean web browsing, video playback, editing documents, etc. You could have a 65W chip, but you don't need extended peak performance. Here's where the current x86 platforms suck. It doesn't have to be, leading many ignorant to believe it has to do with the ISA.

I only believe Lunarlake will do better in this regard because the same insiders that said Meteorlake is a so-so chip isn't saying the same about Lunarlake. I haven't heard of any projections from these people, but they also say Lunarlake isn't plagued with execution issues, delays, and drama like Meteorlake went through.