News Intel Debuts Meteor Lake Die, 'Intel 4' Node: 20% Higher Clocks at Same Power, 2X Area Scaling

[Admin]

Intel debuted the details of its 'Intel 4' process node and shared the first diagram of its Meteor Lake die.

Intel Debuts Meteor Lake Die, 'Intel 4' Node: 20% Higher Clocks at Same Power, 2X Area Scaling : Read more

 

[dehjomz]

Thanks for the write up! After many years of delays, intel will soon use EUV in high volume manufacturing. Looking forward to volume ramp of this node and to final core counts of meteor lake mobile and desktop products.

 

[Jimbojan]

With Meteor Lake, Intel will have a better power efficiency than AMD for the first time in a few years mainly because it is Intel4 line width, which is ahead of TSMC 5nm, where AMD will be using, in addition to better compute performance.

 

[jeremyj_83]

With Meteor Lake, Intel will have a better power efficiency than AMD for the first time in a few years mainly because it is Intel4 line width, which is ahead of TSMC 5nm, where AMD will be using, in addition to better compute performance.

Making this claim with no products out is not possible. Not to mention the fact that Core hasn't been power efficient for over 5 years.

 

[spongiemaster]

Making this claim with no products out is not possible. Not to mention the fact that Core hasn't been power efficient for over 5 years.

Alder Lake is capable of being more power efficient than Zen3. It depends on the application.

Put down your pitchforks: Intel’s 12th-gen CPUs aren’t power hogs

 

[-Fran-]

Alder Lake is capable of being more power efficient than Zen3. It depends on the application.

Put down your pitchforks: Intel’s 12th-gen CPUs aren’t power hogs

Server begs to differ.

EDIT: https://www.anandtech.com/show/16778/amd-epyc-milan-review-part-2/7

And that's not even Milan-X.

Regards.

 

[PCWarrior]

Server begs to differ.

EDIT: https://www.anandtech.com/show/16778/amd-epyc-milan-review-part-2/7

And that's not even Milan-X.

Regards.

You do realise that you are comparing Milan (Zen 3) against Icelake (Sunny Cove) and not Alderlake (Golden Cove) right? And what exactly did you link there? It is not an efficiency chart involving power in any way. All I see is that the EPYC7763 has a lower per thread performance that the Xeon 8380. And the cpus higher in the chart are much lower core-count cpus running at much higher all-core clocks. In any case all these, alone, are not relevant to power/energy efficiency. You need a chart to show power consumption and time to complete a task for that. Also, if you really want to compare apples to apples in terms of efficiency you have to compare cpus of equal core count. Because a higher core count cpu can achieve the same or even higher throughput by running at significantly lower frequencies and thus have significantly lower per core power consumption. Even with the same architecture an 8-core cpu running all cores at 4.8GHz will consume way more power than a 16-core cpu running all cores at 2.4GHz (even if they both complete the task at the same time). That’s because power consumption increases fairly linearly with core count whereas (after a certain frequency) it increases non-linearly (exponentially) with frequency. And, by the way, in idle power comparisons Intel wins comfortably for several generations now (both in equal and non-equal core count comparisons). Especially now with Alderlake and the use of e-cores it is not even close.

 

[-Fran-]

You do realise that you are comparing Milan (Zen 3) against Icelake (Sunny Cove) and not Alderlake (Golden Cove) right? And what exactly did you link there? It is not an efficiency chart involving power in any way. All I see is that the EPYC7763 has a lower per thread performance that the Xeon 8380. And the cpus higher in the chart are much lower core-count cpus running at much higher all-core clocks. In any case all these, alone, are not relevant to power/energy efficiency. You need a chart to show power consumption and time to complete a task for that. Also, if you really want to compare apples to apples in terms of efficiency you have to compare cpus of equal core count. Because a higher core count cpu can achieve the same or even higher throughput by running at significantly lower frequencies and thus have significantly lower per core power consumption. Even with the same architecture an 8-core cpu running all cores at 4.8GHz will consume way more power than a 16-core cpu running all cores at 2.4GHz (even if they both complete the task at the same time). That’s because power consumption increases fairly linearly with core count whereas (after a certain frequency) it increases non-linearly (exponentially) with frequency. And, by the way, in idle power comparisons Intel wins comfortably for several generations now (both in equal and non-equal core count comparisons). Especially now with Alderlake and the use of e-cores it is not even close.

That's pretty much all there is to compare against Intel since they're just not producing proper new server CPUs. They'll be releasing something now, aren't they? And that will be competing with Milan-X and whatever is based off Zen4 now. I can't remember all the names TBH, so sorry for not being super specific.

As for your other points. Who cares about just single performance in servers when you're loading, on average, half the cores on any given workload? Have you monitored server loads at any point in your life? If you have IDLE servers then you should fire whoever handles infrastructure; that's a waste of money. Only exception would be "warm" backups. Everything else "cold" and/or mirrored. IDLE consumption while important for mostly everything, I would not say in servers AMD is not competitive.

Also, I think it's a lesser issue in consumer. I mean, if people can justify the 240W+ consumption of the 12900K at full bore, saying "but the idle power!" is just funny. If you're not using the PC, turn it off. At any other given task, they're close enough to not being an issue and I agree. My point is Enterprise where those overall differences matter and AMD is just better than Intel; sorry you can't see it, but that's just how it is.

Regards.

 

[setx]

Idle power consumption tells absolutely nothing about cores. Because cores at idle consume almost 0. Package consumption is L3 and memory/PCI-E/other controllers.

 

[jeremyj_83]

Alder Lake is capable of being more power efficient than Zen3. It depends on the application.

Put down your pitchforks: Intel’s 12th-gen CPUs aren’t power hogs

Best case I've seen for Alder Lake, 12900k, is when it was limited to 125W. Compared to the 5950X it was still less efficient even though it was limited in power.

 

[JWNoctis]

Couldn't wait until someone chirps in and reiterate how competition is good, and how Zen 3 was price-gouged unobtainium when the best Intel had was Comet Lake. Not that it's untrue, though!

Could remember reading somewhere - sometime in early 2000s, I think - that researchers did not expect much scaling below 90nm. Now there's 5nm node or thereabouts in production, depending on specific marketing terms.

Don't think there'd be much A vs. I arguments left, if 90nm had really turned out to be the practical limit.

 

[funguseater]

I hear Meteor Lake is releasing soon after Star Citizen, what a future to look forward to!

 

[mitch074]

With Meteor Lake, Intel will have a better power efficiency than AMD for the first time in a few years mainly because it is Intel4 line width, which is ahead of TSMC 5nm, where AMD will be using, in addition to better compute performance.

Intel 4 is pretty much equivalent to TSMC 7nm, but maybe a bit more refined. Don't get fooled : up until recently, "Intel 7" was the hated 10 nm. At best, "Intel 4" might equal TSMC's 5nm. But by then, TSMC will have been refining it for several months - better yields. And 3nm will be on the way too.

Alder Lake is capable of being more power efficient than Zen3. It depends on the application.

Put down your pitchforks: Intel’s 12th-gen CPUs aren’t power hogs

That was using Puget System's Adobe benchmarks. Adobe will never optimize their software - they'll simply use Intel's compiler to deliver the x86-64 build and that is all. Thing is, Intel's compiler will stick AMD chips on an unoptimized "generic x86-64" code path and pretty much nerf AMD hardware, sometimes by quite a lot.
Blender scores would be more interesting here, as both chip makers have chipped in (heh) to optimize the software.

 

[spongiemaster]

That's pretty much all there is to compare against Intel since they're just not producing proper new server CPUs. They'll be releasing something now, aren't they? And that will be competing with Milan-X and whatever is based off Zen4 now. I can't remember all the names TBH, so sorry for not being super specific.

As for your other points. Who cares about just single performance in servers when you're loading, on average, half the cores on any given workload? Have you monitored server loads at any point in your life? If you have IDLE servers then you should fire whoever handles infrastructure; that's a waste of money. Only exception would be "warm" backups. Everything else "cold" and/or mirrored. IDLE consumption while important for mostly everything, I would not say in servers AMD is not competitive.

Also, I think it's a lesser issue in consumer. I mean, if people can justify the 240W+ consumption of the 12900K at full bore, saying "but the idle power!" is just funny. If you're not using the PC, turn it off. At any other given task, they're close enough to not being an issue and I agree. My point is Enterprise where those overall differences matter and AMD is just better than Intel; sorry you can't see it, but that's just how it is.

Regards.

Are you in the right thread? Meteor Lake is a consumer architecture. Alder Lake is a consumer architecture. Why are you drifting so far off topic?

 

[spongiemaster]

Intel 4 is pretty much equivalent to TSMC 7nm, but maybe a bit more refined. Don't get fooled : up until recently, "Intel 7" was the hated 10 nm. At best, "Intel 4" might equal TSMC's 5nm. But by then, TSMC will have been refining it for several months - better yields. And 3nm will be on the way too.

That was using Puget System's Adobe benchmarks. Adobe will never optimize their software - they'll simply use Intel's compiler to deliver the x86-64 build and that is all. Thing is, Intel's compiler will stick AMD chips on an unoptimized "generic x86-64" code path and pretty much nerf AMD hardware, sometimes by quite a lot.
Blender scores would be more interesting here, as both chip makers have chipped in (heh) to optimize the software.

Intel 7 is comparable to TSMC 7. That's why Intel renamed it that way.

There were also 2 games in the article. What is the "that doesn't count" excuse for those?

 

[-Fran-]

Are you in the right thread? Meteor Lake is a consumer architecture. Alder Lake is a consumer architecture. Why are you drifting so far off topic?

Because the thread is not just about AL and ML, but the process node. In the servers, where Intel and AMD use the best they have for manufacturing, AMD is ahead right now. I doubt "Intel 4" will make a big difference, but that's just me. The consumer-side AMD SoC is not much more efficient than the monolithic behemoth Intel is using for Alder Lake, but if you look into the Laptop space, where AMD is using a monolithic die, then Intel is at a disadvantage, again. I don't know how else do you want me to explain it?

Regards.

 

[spongiemaster]

I don't know how else do you want me to explain it?

Regards.

You could have just started off with "I'm wrong" and saved a lot of your time. The best of Intel's current Xeon line is using Sunny Cove on Intel's 10nm. Intel has released two architectures since Sunny Cove, Willow Cove and Golden Cove and Alder Lake is using Intel 7 for its node which is a more advanced version of Intel's 10nm than used in current Xeons.
'

 

[Wisecracker]

IEDM 2018: Intel’s 10nm Standard Cell Library and Power Delivery
January 6, 2019

The wheels came off at 14nm. I think it was Kaby - density went craptastic across each cell library and started backing-up. It would be interesting to compare IEDM 2018 cell libraries at 10nm, and what is reported in the "2022 IEEE VLSI Symposium" for 7nm.

 

[-Fran-]

You could have just started off with "I'm wrong" and saved a lot of your time. The best of Intel's current Xeon line is using Sunny Cove on Intel's 10nm. Intel has released two architectures since Sunny Cove, Willow Cove and Golden Cove and Alder Lake is using Intel 7 for its node which is a more advanced version of Intel's 10nm than used in current Xeons.
'

I already provided evidence that AMD is ahead via AnandTech's server tests. Do you need me to Google more for you? I could also get more Laptop data so I can double down on the point, but nah.

The matter of the fact is: Intel even using a more efficient design: monolithic and bigLITTLE is barely on par with AMD's chiplet approach which is inherently less efficient on the desktop and server front. If you don't want to see it, nothing more I can do.

Regards.

 

[JamesJones44]

Intel 4 is pretty much equivalent to TSMC 7nm, but maybe a bit more refined.

Based on the article that doesn't look to be correct. Based on transistor density Intel 4 should closer to TMSC 5nm than 7.

	Intel 4	Intel 7	TSMC N5	TSMC N3
HP Library Density	160 MTr/mm^2 (est.)	80 MTr/mm^2	130 MTr/mm^2 (est.)	208 MTr/mm^2 (est.)

 

[isofilm]

Since Intel 4 will be produced at Fab 34 in Ireland and given Fab 34 received their first UEV tool transplanted from Oregon in April ’22, then keeping in mind that Intel is only receiving 1 EUV tool per month, Fab 34 will not have received all 13 EUV tools required for Meteor Lake until May ’23.

Once the last tool is installed, an additional 3-4 months for calibration, pushes the beginning of Risk Production to 4Q23.

This schedule seems to indicate that Meteor Lake will not start Volume Production in ’23.

If Intel releases Meteor Lake before 2024, it will likely be from low yield (<20%) Risk Production, similar to Samsung 4nm (<20% yield).

 

[SiliconFly]

Making this claim with no products out is not possible. Not to mention the fact that Core hasn't been power efficient for over 5 years.

Intel 4 process is far superior to TSMC N5 & N5P. And meteor lake has already taped-out, engineering sample powered on & tested successfully with windows 11 & ubuntu 22.04. Intel is putting the finishing touches on meteor lake right now. Volume ramp up in Q2 2023.

In other words, Intel takes process leadership over AMD starting next june. And performance leadership too! After that, there's no turning back!

 

[SiliconFly]

To put it mildly, TSMC N5/N5P's slower high density library has LESS density than the faster Intel 4's high performance library!!! This is NO joke guys!!! Meteor lake's Intel 4 node is far superior compared to AMD's TSMC N5/N5P node. Hard facts! Deal with it.

This is gonna be the last year AMD is gonna be on par with intel. Starting next year, AMD will start falling short because of the inherent process advantage intel has in its pipeline. Both homegrown & outsourced. For starters, a very big chunk of TSMC's N3 is now with Intel already (and apple too).

Starting 2023, AMD will be history!

 

[jeremyj_83]

Intel 4 process is far superior to TSMC N5 & N5P. And meteor lake has already taped-out, engineering sample powered on & tested with windows 11 & ubuntu 22.04. Intel is putting the finishing touches on meteor lake right now. Volume ramp up in Q2 2023.

In other words, Intel takes process leadership over AMD starting next june. After that, there's no turning back!

They are not going to be doing volume ramp in Q2 for Meteor Lake when Raptor Lake won't be shipping until H2 this year. At best it will be a 12 month cadence which puts Meteor Lake in H2 2023.

Who says that Intel 4 is superior to TSMC N5? Intel, you, who?

 

[SiliconFly]

They are not going to be doing volume ramp in Q2 for Meteor Lake when Raptor Lake won't be shipping until H2 this year. At best it will be a 12 month cadence which puts Meteor Lake in H2 2023.

Who says that Intel 4 is superior to TSMC N5? Intel, you, who?

Tom's hardware