Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

Intel confirms next-gen 15th Gen Core 'Arrow Lake' CPUs for 2024

Intel's new 15th Gen Core 'Arrow Lake' CPUs will be made on Intel 4, Intel 20A, and external N3 nodes when it drops in 2024.

www.tweaktown.com

Take that with a huge rock of salt 🙂

By the way that article is misunderstanding Intel's presentation.

Intel 4 is for Meteorlake, 20A is for Arrowlake, and TSMC 3 is for the GPU tiles, likely on both.

 

[Accord99]

There is probably something wrong there with the Intel system, because Golden Cove is EXCEPTIONALLY strong in Cinebench.

For 104 GC cores, they must be running at <=2GHz to get such a low score.

 

[Markfw]

For 104 GC cores, they must be running at <=2GHz to get such a low score.

Its possible they will be slow, to fit in a 370 watt envelope. Thats 53 watts per 8 cores, less than any retail Alderlake. And thats 56 cores, they may be at 52.

At any rate, we have to wait for official reviews, but the one above looks pretty close, and it says Sapphire rapids will get killed in performance comparisons and generate a lot of heat and use a lot of power. Read the clip he copied from that review.

Edit: I just copied it


"Although this is an ES chip and we expect performance to improve significantly for the Sapphire Rapids-SP chip in the QS state, the final chip may still be only competitive against AMD's EPYC Milan parts while it will be launching at the same time when AMD will come out guns blazing with its next-generation EPYC Genoa 7004 CPUs. With such high power draw and not enough performance at hand, it looks like Intel will be getting a serious beating in the server segment by AMD, something that has been the case over the past few generations"

 

[DrMrLordX]

Arrow Lake is a 2024 platform confirmed by Intel in February, it won't or cannot come before mid 2024 because of 20A which is manufacturing ready in H1 2024.

Oh right they're still claiming Arrow Lake will be on 20a in 2024. Color me skeptical. Meanwhile they're taking huge allocations of N3. They're going to do something with it, and N3 will be ready well before 20a.

I would be very surprised if all that N3 went only to Arc GPUs and GPU tiles.

I wouldn't read too much into SPR supporting 4S and 8S. Intel has always supported those in the past.

They have supported them, but only in very limited volume recently (see: Cooper Lake).

 

[jpiniero]

They have supported them, but only in very limited volume recently (see: Cooper Lake).

There's a limited demand for 4S and 8S so it checks out. The big time gap between Cooper and Sapphire is pretty obviously because they held Sapphire back (and Icelake Server) because of 10 nm yield.

 

[Exist50]

8S and quite possible 4S servers are probably not going to be around past SPR/EMR. What's the point when you can get 100-200 cores and 12 memory channels per socket?

My expectation is that anything that uses I4/I3/20A is going to be tiny volume at best, with the hope that they can make 18A work eventually.

Nah, they should at least get high volume on I3. I think 20A will depend on timing. If they actually manage to hold to their roadmap, 20A would only be around long enough for some ARL tiles. But if 18A got delayed, they might try to make a server chip with 20A.

 

[jpiniero]

8S and quite possible 4S servers are probably not going to be around past SPR/EMR. What's the point when you can get 100-200 cores and 12 memory channels per socket?

The software they run always wants more and more... Power consumption might be the bigger deal.

 

[JoeRambo]

There is probably something wrong there with the Intel system, because Golden Cove is EXCEPTIONALLY strong in Cinebench.

Understatement of the week. 104 cores of ADL scoring 80K in CB23 at 2x370W power taken as gospel is hilarious.

 

[Markfw]

Understatement of the week. 104 cores of ADL scoring 80K in CB23 at 2x370W power taken as gospel is hilarious.

No, these are not retail, ES chips, certainly possible.

 

[JoeRambo]

No, these are not retail, ES chips, certainly possible.

That's why i have been warning not to put any weight into those scores, yet You are linking to this thread and making bold claims as if it was ground truth.

And as for how well its doing ? see here, it will be crushed:

 

[lobz]

That's why i have been warning not to put any weight into those scores, yet You are linking to this thread and making bold claims as if it was ground truth.

Read again. He meant by Genoa in the server space. He's also 99,5% surely right about that.

 

[JoeRambo]

Read again. He meant by Genoa in the server space. He's also 99,5% surely right about that.

So basically comparing a product we have no reliable scores, using a proxy of claims about other product in who knows what configuration using pre-release silicon? Anandtech forums in nutshell.

While i have great hopes for Zen4, i will wait for someone like ServeTheHome and Phoronix to help me form the informed opinion, not some blatantly obvious wrong score.

 

[coercitiv]

Understatement of the week. 104 cores of ADL scoring 80K in CB23 at 2x370W power taken as gospel is hilarious.

Pretty much. Basic math using CB23 performance data for Golden Cove says an 80K score is what a single SPR 52 core SKU @ 370W should be able to achieve.

If we ignore uncore power on both SPR and ADL, 52 cores @ 370W results in around 7.1W / core. Punch in this number to obtain a normalized power baseline for ADL 8c/16t, and you get ~56W. For reference ADL-S 8c/16t does 14K score in CB23 with an average package power of 62W. Even if we lower the score to around 12K to account for much higher uncore power in SPR than a linear scaling of 6x from ADL-S, we can still confidently say 8 SPR cores can score 12K+ in CB23, which means 78K+ for 52 cores @ 370W running somwhere around 3.5Ghz.

@nicalandia is simply echoing rumor mill data without bothering to double check, and all we get here is poor quality discussions from people who are just too lazy to punch in the numbers.

Not every rumor is worth chewing!

 

[eek2121]

Oh right they're still claiming Arrow Lake will be on 20a in 2024. Color me skeptical. Meanwhile they're taking huge allocations of N3. They're going to do something with it, and N3 will be ready well before 20a.

I would be very surprised if all that N3 went only to Arc GPUs and GPU tiles.



They have supported them, but only in very limited volume recently (see: Cooper Lake).

TSMC is being used for the GPUs, of course they are going to have tons of allocation. I suspect some of you don’t realize how many CPUs Intel sells every year. If each CPU counts as a GPU (since most have integrated GPUs), Intel sells more GPUs in a given year than NVIDIA or AMD.

 

[DrMrLordX]

TSMC is being used for the GPUs, of course they are going to have tons of allocation. I suspect some of you don’t realize how many CPUs Intel sells every year. If each CPU counts as a GPU (since most have integrated GPUs), Intel sells more GPUs in a given year than NVIDIA or AMD.

We're talking 40 kwpm of N3. AMD's entire allocation of N5, at least initially, is alleged to be ~20 kwpm. Intel doesn't ship THAT many iGPUs. I don't think their entire 14nm production run got over 80 kwpm. And they used that for everything.

 

[lobz]

That's why i have been warning not to put any weight into those scores, yet You are linking to this thread and making bold claims as if it was ground truth.

This score had nothing to do with him thinking that spr will be crushed by genoa. Quite the trip you're taking there.

 

[eek2121]

We're talking 40 kwpm of N3. AMD's entire allocation of N5, at least initially, is alleged to be ~20 kwpm. Intel doesn't ship THAT many iGPUs. I don't think their entire 14nm production run got over 80 kwpm. And they used that for everything.

Every single CPU they make has a GPU. They are also betting big on their discrete GPUs.

The numbers don’t seem out of line to me. I can easily see that many GPUs being shipped. Note that future integrated GPUs are speculated to be larger than the current ones.

 

[dullard]

First I'm hearing of that; granted, Meteor Lake was supposed to come out in 2022 and Arrow Lake in 2023. They could push Arrow back to 2024.

How can that be the first you are hearing that? Every rumor for years was Meteor Lake being 14th gen and Arrow Lake being 15th gen. The only thing up in the air was some people here claimed that Meteor Lake was mobile only despite rumors otherwise.

 

[jpiniero]

How can that be the first you are hearing that? Every rumor for years was Meteor Lake being 14th gen and Arrow Lake being 15th gen. The only thing up in the air was some people here claimed that Meteor Lake was mobile only despite rumors otherwise.

Unless the CPU chiplet is being fabbed at TSMC there is no way there will be a desktop version. They can't hide the volume like they can on mobile.

 

[dullard]

Unless the CPU chiplet is being fabbed at TSMC there is no way there will be a desktop version. They can't hide the volume like they can on mobile.

Meteor Lake (Intel 4) only uses a bit of EUV. Much of it will be on the same type of equipment they are already using for Alder Lake. It is Intel 3 with much more EUV needed where they really don't have the capacity to fab enough.

Here are some desktop Meteor Lake rumors:

• Mar 23, 2021: https://videocardz.com/newz/intels-first-7nm-client-architecture-meteor-lake-to-launch-in-2023 "Meteor Lake is Intel’s first 7nm client architecture, which means it will launch for desktop and mobile users under the Intel Core series." Note: Intel has since switched this naming to Intel 4 instead of 7 nm.
• May 24, 2021: https://videocardz.com/newz/intels-7nm-meteor-lake-compute-tile-has-taped-in "The series is rumored to be the last to support the LGA1700 socket" This LGA1700 socket is a desktop socket.
• May 26, 2021: https://www.techadvisor.com/news/pc-components/intel-meteor-lake-14th-gen-cpu-3803533/ "Will Intel release Meteor Lake desktop and mobile chips? Most likely, although their releases may be staggered".
• July 26, 2021: https://videocardz.com/newz/intel-r...re-rapids-meteor-lake-and-granite-rapids-cpus "Intel confirmed that Meteor Lake will use Foveros packaging technology and it will support TDP ranges from 5 to 125W."
  • This 125 W is direct from Intel itself, slide 35: https://download.intel.com/newsroom/2021/client-computing/Intel-Accelerated-2021-presentation.pdf There are not that many 125 W CPUs that are mobile only.
• Nov 18, 2021: https://videocardz.com/newz/intel-14th-gen-core-meteor-lake-mobile-test-cpu-has-been-pictured "Intel Meteor Lake desktop and mobile CPU series should not be expected till the second quarter of 2023."
• Feb 18, 2022: https://wccftech.com/intel-next-gen...gen-arrow-lake-2024-16th-gen-lunar-lake-2025/ "Intel Confirms Raptor Lake, Meteor Lake, Arrow Lake, Lunar Lake Client Desktop CPU Platforms"
• Apr 28, 2022: https://videocardz.com/newz/14th-gen-core-meteor-lake-has-been-powered-on-on-track-to-launch-in-2023 "Therefore, Meteor Lake will debut on low-power laptops as well as high-end desktop consumer systems."

Of course those are mostly rumors or forward-looking statements from Intel. Thus things in the future can change. But for that to be DrMrLordX's first time hearing of desktop Meteor Lake meant his eyes were closed to the rumors.

 

[igor_kavinski]

• May 24, 2021: https://videocardz.com/newz/intels-7nm-meteor-lake-compute-tile-has-taped-in "The series is rumored to be the last to support the LGA1700 socket" This LGA1700 socket is a desktop socket.

So Meteor Lake will have the same socket as ADL/RPL but still require new mobos?

Great work chronicling the rumors, by the way.

 

[jpiniero]

Mar 23, 2021: https://videocardz.com/newz/intels-first-7nm-client-architecture-meteor-lake-to-launch-in-2023 "Meteor Lake is Intel’s first 7nm client architecture, which means it will launch for desktop

No it doesn't. Client is a superset of desktop and mobile.

This 125 W is direct from Intel itself, slide 35: https://download.intel.com/newsroom/2021/client-computing/Intel-Accelerated-2021-presentation.pdf There are not that many 125 W CPUs that are mobile only

Alder Lake H line is 115.

All of the other links doesn't quote Intel at all.

 

[dullard]

Alder Lake H line is 115.

All of the other links doesn't quote Intel at all.

Alder Lake H is 45 W, boost 115 W. That is a completely different thing than a 125 W chip.

Why does it matter if they quote Intel for this discussion? The statement is that he never heard of it. All links are to ways he could have heard of it. The statement in discussion was not "is desktop Meteor Lake guaranteed to exist", it was about whether or not there was even a way to have heard about it.

Edit: Intel Chromium OS driver mentions Meteor Lake-S (-S is the suffix for desktop processors): https://github.com/intel/dptf/blob/...435925407c0/Common/esif_ccb_cpuid.h#L108-L112

#define CPUID_FAMILY_MODEL_MTL_S 0x000606C0 // Meteor Lake S

 

[mikk]

Unless the CPU chiplet is being fabbed at TSMC there is no way there will be a desktop version. They can't hide the volume like they can on mobile.

Maybe it's a limited desktop launch (in 2023). Like ADL-S being K-SKU limited in 2021. A full scale desktop lineup is unlikely because of the volume (EUV) limitation in 2023.

Why does it matter if they quote Intel for this discussion? The statement is that he never heard of it. All links are to ways he could have heard of it. The statement in discussion was not "is desktop Meteor Lake guaranteed to exist", it was about whether or not there was even a way to have heard about it.

Depends on the source, most of your links are irrelevant, they have no source. There are only two reliable sources for a desktop version as far as I know, the one where Intel told us it will scale from 5-125W and the second is MLID, he claimed there will be a desktop version.

 

[dullard]

Depends on the source, most of your links are irrelevant, they have no source. There are only two reliable sources for a desktop version as far as I know, the one where Intel told us it will scale from 5-125W and the second is MLID, he claimed there will be a desktop version.

You still are missing my point. You are trying to prove if desktop meteor lake exists or not. That is not what I was discussing. I was discussing the fact that he claims that never heard of the concept of a desktop meteor lake.

As for EUV, meteor lake barely uses Intel's EUV. Just on the CPU tile and just on a part of the back end of line of that tile. When it comes to Intel 3, then there is no way to have a full lineup due to much increased EUV use. That lead to the rumors of server chips only, use of TSMC for the CPU, etc.