Discussion Intel current and future Lakes & Rapids thread

[SpudLobby]

Who won that socket that Qualcomm lost

Any guesses as to who won the socket that Qualcomm lost?

semiaccurate.com

This to me seems like it’s about PCs, the way he says socket and of course it’s Charlie. Wonder if Intel won some laptop from Dell (but that doesn’t make sense, XPS is Intel exclusively), maybe MediaTek? But Intel in some low power domain makes sense.

The phrasing however is odd.

“SemiAccurate wasn’t expecting this name to pop up on the list but it makes sense given the company’s ambitions.”

Nvidia or MediaTek to me seem most likely from this phrasing.

Reflexively from the title alone I would think this is about Lunar Lake or ARL beating Qc, but I wouldn’t bet on it, and feel it may be one of those two above. It is about mobile PC’s though imo.

 

[mikk]

Nah they’ll use N3E.

Why do you think so? Everyone says it's N3B.

As for DT ARL it is supposed to be fabbed with Intel 4 or something like this, so it should scale differently than a TMSC s fabbed CPU, perhaps that the mobile variant, wich require eventualy greater efficency, could be fabbed at TSMC.

Intel 4 what? It's N3 or 20A. At the moment it seems like only the 6+8 tile uses N20A whereas 8+16 tile uses N3 on ARL-S.

 

[Abwx]

Intel 4 what? It's N3 or 20A. At the moment it seems like only the 6+8 tile uses N20A whereas 8+16 tile uses N3 on ARL-S.

It would make sense to use N3 where very good perf/watt is necessary, that is for mobile parts.

As for DT ARL N3 should had yielded much better perf/watt improvement than what is displayed in the leaked slides.

N5 would be enough to reach such numbers, and since Intel 4 shouldnt be far from N5 it would be more relevant to use this in house process, but we ll see once we have more precise infos.

 

[Geddagod]

It would make sense to use N3 where very good perf/watt is necessary, that is for mobile parts.

As for DT ARL N3 should had yielded much better perf/watt improvement than what is displayed in the leaked slides.

N5 would be enough to reach such numbers, and since Intel 4 shouldnt be far from N5 it would be more relevant to use this in house process, but we ll see once we have more precise infos.

Literally everyone claims ARL is going to be on 20A and TSMC 3nm. Intel themselves confirmed 20A, and I think it's heavily, heavily implied that TSMC 3nm is also being used for ARL. No node info for ARL is conflicting, other than whether ARL killed off the 20A variant, but everyone agrees at this point it's 3nm for ARL.
Idk why people are pushing that ARL/LNC would/should be on Intel 3/4. Both 20A/TSMC 3nm variants being equally hitting below targets suggest it's a fundamental architectural problem or high level layout of the core. If the problem is ST freq max being absurdly low (think ~5ghz) and it's mostly because the node isn't mature enough, then perhaps it's reasonable to move it back a node (though obv that's gonna be delayed)....
But then we also have to think about the core area. Based on the entire core complex (including ringbus+l3) looking to be around the same size as RWC, moving back a node has the added effect of significant increasing the area of the core, and other complexities added with backporting. It honestly might not be worth it.

 

[Abwx]

Literally everyone claims ARL is going to be on 20A and TSMC 3nm. Intel themselves confirmed 20A, and I think it's heavily, heavily implied that TSMC 3nm is also being used for ARL. No node info for ARL is conflicting, other than whether ARL killed off the 20A variant, but everyone agrees at this point it's 3nm for ARL.
Idk why people are pushing that ARL/LNC would/should be on Intel 3/4. Both 20A/TSMC 3nm variants being equally hitting below targets suggest it's a fundamental architectural problem or high level layout of the core. If the problem is ST freq max being absurdly low (think ~5ghz) and it's mostly because the node isn't mature enough, then perhaps it's reasonable to move it back a node (though obv that's gonna be delayed)....
But then we also have to think about the core area. Based on the entire core complex (including ringbus+l3) looking to be around the same size as RWC, moving back a node has the added effect of significant increasing the area of the core, and other complexities added with backporting. It honestly might not be worth it.

You miss the point, doesnt matter what everyone is saying, AMD s 5N fabbed 7950X has 14% higher perf than the 13900K at isopower, or said otherwise 30% lower power at isoperf.

If Intel was to use N5 for RPL they would get similar numbers as AMD, and with N3 it would be quite more perfs (or lower power) than with N5, and this apply to DT ARL as well, yet the leaked slides say otherwise, the improvement is about the one that would be brought by using N5, so i dont think that DT ARL use N3 but rather Intel 4 wich should be close to N5.

Now possibly that the mobile dedicated SKUs use N3, hence a confusion about DT ARL also using this process.

 

[Exist50]

I should just ignore it, but I can't believe you all even bother to comment on MLID - I am assuming no one watches other than as satire. Yeah, he's an influential moron so I'm sure that gets old what with the rumors he promulgates, but you really don't need to watch to refute every last bullet point like it's a Snopes C-tier takedown.

IMO: Bepo is unintentionally funny in a way MLID could only dream of.

Yeah, you have a point. Taking the bait, as it were. Though I do think it's important for anyone knew to the forum to understand that these "leaks" are not to be taken seriously.

 

[SpudLobby]

Yeah, you have a point. Taking the bait, as it were. Though I do think it's important for anyone knew to the forum to understand that these "leaks" are not to be taken seriously.

Yeah, at risk of being rude I should specify that I meant this broadly to a degree and not really as condescension narrowly directed to the forum participants here. On Twitter, Reddit or elsewhere relatively sane, smart people give this guy what he wants too often, it sincerely reminds me of the lowest forms of political drivel. That said, there are standards and at some point you have to address this stuff, just saying it feels out of control, lol.

 

[SpudLobby]

Why do you think so? Everyone says it's N3B.

I suppose I thought they'd like to make use of FinFlex which offers much more granular + better yields/lower cost in N3E and assumed ARL wouldn't really come till mid 2025 at this point so why not. N3B to me (besides previous rumors of an N3B MTL iGPU etc) felt more like an Apple thing given costs and maturity.

Does it really matter? Maybe not, but.

 

[eek2121]

In these slides ARL stated perf is at a stated 250W, how could it achive those perfs at 120-150W.?..

253W for RPL and 250W for ARL :

Intel Arrow Lake-S to be 6% to 21% faster than Raptor Lake-S, Xe-LPG graphics show up to 2.4x improvement - VideoCardz.com

Intel Arrow Lake-S up to 21% faster than Raptor Lake CPU Igor’sLAB recently obtained an internal presentation from Intel revealing performance projections for the upcoming Arrow Lake-S, the next-generation client platform. Although the platform’s launch is still some time away, the discussed...

videocardz.com

I'm willing to bet that slide was taken completely out of context with regards to the power numbers. I don't know the source, but this slide just screams someone explaining "ARROW LAKE DOESN'T SCALE WITH A HIGHER PL2 SO WE SHOULDN'T DO IT" or something similar. It could also be an extremely old/out of date slide. We don't know.

It's silly for people to be blasting a product based on unverified rumors and leaks. Just wait for something concrete, like actual performance leaks.

Meteor Lake will have no issue keeping up with Raptor Lake, despite having lower power consumption, and it is on Intel 4. Arrow Lake will be far more efficient than Meteor Lake.

The other thing that bugs me is Intel is supposedly adding 2 LP cores to the SOC tile for Meteor Lake (unconfirmed), but suddenly they are gone again for Arrow Lake?

 

[Geddagod]

You miss the point, doesnt matter what everyone is saying, AMD s 5N fabbed 7950X has 14% higher perf than the 13900K at isopower, or said otherwise 30% lower power at isoperf.

If Intel was to use N5 for RPL they would get similar numbers as AMD, and with N3 it would be quite more perfs (or lower power) than with N5, and this apply to DT ARL as well, yet the leaked slides say otherwise, the improvement is about the one that would be brought by using N5, so i dont think that DT ARL use N3 but rather Intel 4 wich should be close to N5.

Now possibly that the mobile dedicated SKUs use N3, hence a confusion about DT ARL also using this process.

Intel has repeatedly claimed they are using 20A for ARL. I don't know what else to tell you.

What do you think is more likely:
Intel has a secret Intel 4 ARL sku, along with TSMC 3nm and also Intel 20A, or...
Intel missed design targets on a riskier overhauled core on immature processes

Also you can't use that chart to deduce ARL performance across the curve, who knows how it scales?

 

[jpiniero]

Intel has repeatedly claimed they are using 20A for ARL. I don't know what else to tell you.

What do you think is more likely:
Intel has a secret Intel 4 ARL sku, along with TSMC 3nm and also Intel 20A, or...
Intel missed design targets on a riskier overhauled core on immature processes

Also you can't use that chart to deduce ARL performance across the curve, who knows how it scales?

It was probally dual sourced between 20A and N3E, and they cancelled the 20A. That seems the most likely outcome.

 

[Geddagod]

I'm willing to bet that slide was taken completely out of context with regards to the power numbers. I don't know the source, but this slide just screams someone explaining "ARROW LAKE DOESN'T SCALE WITH A HIGHER PL2 SO WE SHOULDN'T DO IT" or something similar. It could also be an extremely old/out of date slide. We don't know.

It's silly for people to be blasting a product based on unverified rumors and leaks. Just wait for something concrete, like actual performance leaks.

Meteor Lake will have no issue keeping up with Raptor Lake, despite having lower power consumption, and it is on Intel 4. Arrow Lake will be far more efficient than Meteor Lake.

The other thing that bugs me is Intel is supposedly adding 2 LP cores to the SOC tile for Meteor Lake (unconfirmed), but suddenly they are gone again for Arrow Lake?

Apparently it's a very recent slide.
Based on the consensus from other leakers, this appears to be true, and if it's true, people should be blasting Intel for the mediocre gain in performance. Also actual performance leaks don't start appearing till like a couple months before launch :/
I don't think too many people are doubting ARL efficiency being higher than MTL, it's just the perf....
I don't think the 2LP cores are gone from ARL? Was that in the igor leak or something? LP cores are rumored to be removed from LNL tho.

 

[Geddagod]

It was probally dual sourced between 20A and N3E, and they cancelled the 20A. That seems the most likely outcome.

You can't cancel and port a product onto a new node this 'early' (relatively speaking ofc) before launch. There quite simply is no time for it. Gelsinger talked about 20A for ARL in Q1 2023. Setting them back all the way to design complete/just before tape in, at this point in time, means that they will be roughly an year behind schedule.

 

[jpiniero]

You can't cancel and port a product onto a new node this 'early' (relatively speaking ofc) before launch. There quite simply is no time for it. Gelsinger talked about 20A for ARL in Q1 2023. Setting them back all the way to design complete/just before tape in, at this point in time, means that they will be roughly an year behind schedule.

If I wasn't clear, I mean the plan was to do (one of?) the Arrow mobile CPU tiles on both N3E and 20A. And then they decided to just cancel the 20A one.

 

[H433x0n]

Apparently it's a very recent slide.
Based on the consensus from other leakers, this appears to be true, and if it's true, people should be blasting Intel for the mediocre gain in performance. Also actual performance leaks don't start appearing till like a couple months before launch :/
I don't think too many people are doubting ARL efficiency being higher than MTL, it's just the perf....
I don't think the 2LP cores are gone from ARL? Was that in the igor leak or something? LP cores are rumored to be removed from LNL tho.

My concern is that if those numbers are from an early stepping clocking at 5ghz, the arch is pretty much boned. For it to be redeemable the results from SpecInt and Specfp would have had to be ran at 4.5-4.6ghz. That would at least put ST performance >10% from RPL-R.

 

[Geddagod]

My concern is that if those numbers are from an early stepping clocking at 5ghz, the arch is pretty much boned. For it to be redeemable the results from SpecInt and Specfp would have had to be ran at 4.5-4.6ghz. That would at least put ST performance >10% from RPL-R.

ARL looks like a dud, onto PTL cope, and hopes cougar cove fixes whatever problems lion cove has.
jk

 

[DrMrLordX]

if it's true, people should be blasting Intel for the mediocre gain in performance.

Those numbers would be acceptable for a hypothetical Meteor Lake-S 8+16 chip that will never exist. In fact that's what most optimists would have expected as far out as a year ago. Kinda makes you wonder . . .

I don't think too many people are doubting ARL efficiency being higher than MTL, it's just the perf....

It does say that the listed part is 250W PL2. It looks like it may be way out of its efficiency range.

 

[S'renne]

Is there anything on ARL iGPU btw..

 

[Geddagod]

Is there anything on ARL iGPU btw..

2.1-2.4x RPL-S iGPU perf in 3Dmark benches.

 

[S'renne]

2.1-2.4x RPL-S iGPU perf in 3Dmark benches.

What's it supposed to be equivalent to, 780M? Also thanks...would be really great if iGPU gets a huge upgrade after Arrow Lake too

 

[mikk]

What's it supposed to be equivalent to, 780M? Also thanks...would be really great if iGPU gets a huge upgrade after Arrow Lake too

No it's not even close because the baseline is RPL-S GT1 32EU versus ARL-S GT1 64EUs. RPL-S iGPU gets 1000 points in 3dmark timespy and with the ARL-S projection it can reach 2300-2400 points. It's faster than a current Xe LP GT2 with 96EUs though, for a desktop CPU not so bad. For mobile they have a GT2 version and maybe even GT3.

 

[SiliconFly]

You can't cancel and port a product onto a new node this 'early' (relatively speaking ofc) before launch. There quite simply is no time for it. Gelsinger talked about 20A for ARL in Q1 2023. Setting them back all the way to design complete/just before tape in, at this point in time, means that they will be roughly an year behind schedule.

Any news on the health of Intel 20A? When it comes to 20A, the silence is eerie! Hope they shed some light sooner rather than later

 

[eek2121]

Any news on the health of Intel 20A? When it comes to 20A, the silence is eerie! Hope they shed some light sooner rather than later

We will probably find out later this week.

 

[dullard]

Any news on the health of Intel 20A? When it comes to 20A, the silence is eerie! Hope they shed some light sooner rather than later

We had a 20A newsroom dump from Intel on June 5.

With PowerVia, Intel Achieves a Chipmaking Breakthrough - Intel Newsroom

Intel’s Ben Sell explains how the company developed and proved the world's first backside power solution, taking a major step forward in chipmaking.

www.intel.com

PowerVia Test Shows Industry-Leading Performance - Intel Newsroom

Intel is first to implement backside power in a product-like chip, resulting in over 90% cell utilization and other gains.

www.intel.com

Being marketing pieces, take those with as large a grain of salt as you wish. But they claim "PowerVia, which will be introduced on the Intel 20A process node in the first half of 2024". I'm not sure that I'd call that eerie silence.

As eek2121 mentioned, we will get an Intel Q2 2023 financial results on this Thursday. That might include more 20A.

As for "health" I guess the best we can go off of is this chart:

Once we see Intel 4 this Fall, we'll have an idea of 20A being ~2 quarters behind in yield.

 

[H433x0n]

Any news on the health of Intel 20A? When it comes to 20A, the silence is eerie! Hope they shed some light sooner rather than later

They had a lot of info for VLSI a few months ago on backside power delivery (PowerVia). PowerVia is the only real public disclosure on 20A/18A progress.

The Intel 3, 20A and 18A nodes have had white papers published for industry insiders but sparse public information.

They‘ll probably publicly state that Intel 4 is in HVM and give info on Intel 3 in the Q2 earnings. I doubt we’ll get much detailed info for 20A/18A, I wouldn’t expect much for another 6 months.