Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

Raichu/kopite7kimi: Arrowlake-S high end, Meteorlake-S low-mid.

According to Raichu and kopite7kimi MTL-S will be a mid to lowend Mainstream series for desktop while ARL-S is coming for the highend.

Raichu says MTL-S and ARL-S will come almost at the same time. Or it means MTL-S is more towards H2 2024 as well.

Raichu: MTL-S and ARL-S coming almost the same time.

Of course, that means Intel will be competing with reheated Alder Lake all the way till some indeterminate time in 2024, which is a terrible position indeed. Even when Arrow Lake comes,

kopite7kimi: Meteorlake-S in Q1 2024.

Doesn't seem like terrible at all to me. This is just as I was speculating. They are doing this to speed up the process launch. I assume they will reduce the gap between process readiness and product releases as well.

 

[nicalandia]

kopite7kimi: Meteorlake-S in Q1 2024.

Doesn't seem like terrible at all to me. This is just as I was speculating.

That's Terrible for Xeons..... Arrow Lake/Granite Rapids are arriving by Q2 2024 if no further delays.

 

[IntelUser2000]

That's Terrible for Xeons..... Arrow Lake/Granite Rapids are arriving by Q2 2024 if no further delays.

Xeons were always late because of validation required. I never believed them when they said Xeons would get the first process. Yea sure, good luck. The same applies for exotic memory products like ones with eDRAM, or even AMD's VCache chips. Same will apply to Arrowlake with 384EUs and ADM technology.

Although there's still an opportunity to speed things up in Xeons as well.

Sapphire Rapids 4S+ - Q4 2022
Sapphire Rapids 2S - Q1 2023?

Only Emerald Rapids showing in leaked roadmaps is 4S+. Yes it could be that 2S isn't showing and coming later. But they have a desire to speed things up and a 64 core server for end 2023 is... not desirable to put it lightly.

Speculation-
4S+ = Emerald Rapids: Q4 2023
2S = Sapphire Rapids with price reduction

4S+ = Granite Rapids: Q4 2024
2S = Granite Rapids Q1/Q2 2024

This will also interestingly harmonize with how previously the Xeon MPs/Xeon E7s came later than Xeon DPs/Xeon E5s. The enterprise goes Intel anyways with the 4S+ option so they have less competition there. AMD has EPYC for 2S so that's where they really need to worry.

 

[DrMrLordX]

According to Raichu and kopite7kimi MTL-S will be a mid to lowend Mainstream series for desktop while ARL-S is coming for the highend. I believe them. MTL-S 6+8 maybe the maximum?

To date we haven't seen any Meteor Lakes shown off that would indicate anything other than a mobile die, and mobile should be 2P+8e -> 6p+8e. They could easily slap that in an LGA form factor and call it a day. Not impressive, but doable.

Difficult for Desktop and Plain Awful for Data Center/Severs....

Meteor Lake and Arrow Lake have nothing to do with datacentre products. Granite Rapids has been pushed back to 2024 on its own separate timeline.

 

[nicalandia]

Meteor Lake and Arrow Lake have nothing to do with datacentre products. Granite Rapids has been pushed back to 2024 on its own separate timeline.

All I was saying is that Intel will have to fight AMD on both High End and Data Center with Golden Cove/Raptor Cove(both 10nm+) until Lion Cove(Arrow Lake and Granite Rapids) shows up at Q1/Q2 of 2024....


That's a bleak future.

From
96C/192T Genoa vs 56C/112T Sapphire Rapids in 2022
128C/256T Bergamo vs 64C/128T Emerald Rapids in 2023

 

[Exist50]

I hope intel has a backup plan for arrow lake if 20A hits any trouble. maybe they can backport Lion cove and Skymont to intel 3 if needed.

That basically sounds like what Granite Rapids/Sierra Forest are, so it should be doable. Or there's N3.

No Chance AMD is using N3 from TSCM before Apple, they are schedule to mass produce N3 late 2023 and nowing AMD Process node and release cycle (Zen/Zen+ using 14nm/12nm, Zen2/Zen3 using N7, we can expect Zen4/Zen5 to be on N5)

What? N3 is supposed to be volume production this year, not late next.

 

[igor_kavinski]

That's a bleak future.

From
96C/192T Genoa vs 56C/112T Sapphire Rapids in 2022
128C/256T Bergamo vs 64C/128T Emerald Rapids in 2023

Nope. Not bleak at all. It makes Intel look bad but AMD can't provide enough chips for all enterprise customers so Intel still sells a lot of server chips despite having fewer cores. Their customers will just buy double for double the amount of cores. What else can they do if AMD can't meet demand?

 

[nicalandia]

What? N3 is supposed to be volume production this year, not late next.

"Apple is expected to release the majority of its devices with 3nm chips fabricated by TSMC in 2023"

TSMC on Track to Start 3nm Chip Production in Second Half of This Year

Apple chipmaking partner TSMC says it will be ready to move its 3nm chip process to volume production in the second half of this year, putting it on...

www.macrumors.com

You can bet AMD will not have N3 chips ready by 2023

 

[Exist50]

Xeons were always late because of validation required. I never believed them when they said Xeons would get the first process. Yea sure, good luck. The same applies for exotic memory products like ones with eDRAM, or even AMD's VCache chips. Same will apply to Arrowlake with 384EUs and ADM technology.

Although there's still an opportunity to speed things up in Xeons as well.

Sapphire Rapids 4S+ - Q4 2022
Sapphire Rapids 2S - Q1 2023?

Only Emerald Rapids showing in leaked roadmaps is 4S+. Yes it could be that 2S isn't showing and coming later. But they have a desire to speed things up and a 64 core server for end 2023 is... not desirable to put it lightly.

Speculation-
4S+ = Emerald Rapids: Q4 2023
2S = Sapphire Rapids with price reduction

4S+ = Granite Rapids: Q4 2024
2S = Granite Rapids Q1/Q2 2024

This will also interestingly harmonize with how previously the Xeon MPs/Xeon E7s came later than Xeon DPs/Xeon E5s. The enterprise goes Intel anyways with the 4S+ option so they have less competition there. AMD has EPYC for 2S so that's where they really need to worry.

I'm pretty sure that when they list a number of sockets, it's an "up to" number. 4S/8S being harder to support than 2S.

 

[Exist50]

"Apple is expected to release the majority of its devices with 3nm chips fabricated by TSMC in 2023"

TSMC on Track to Start 3nm Chip Production in Second Half of This Year

Apple chipmaking partner TSMC says it will be ready to move its 3nm chip process to volume production in the second half of this year, putting it on...

www.macrumors.com

You can bet AMD will not have N3 chips ready by 2023

Ok? N3 slipped Apple's biggest launch window. Doesn't mean no one else could or will use it. The contrary, if anything.

 

[moinmoin]

Ok? N3 slipped Apple's biggest launch window. Doesn't mean no one else could or will use it. The contrary, if anything.

I'm with @nicalandia there, personally I don't expect AMD to use N3 before Zen 6.

 

[nicalandia]

Nope. Not bleak at all. It makes Intel look bad

I guess this turned out to be true..

Intel CEO's Comments of AMD 'in the Rearview Mirror' May Indicate Intel is Driving Backwards: Analyst | Hardware Times

A while back Intel CEO Pat Gelsinger claimed that the chipmaker was back in the game, and AMD is (and will continue to be) in the rear-view mirror, and never again in the windshield. Some may call these bold claims of supposedly retaining market share Intel’s newfound confidence, but as...

www.hardwaretimes.com

 

[uzzi38]

All I was saying is that Intel will have to fight AMD on both High End and Data Center with Golden Cove/Raptor Cove(both 10nm+) until Lion Cove(Arrow Lake and Granite Rapids) shows up at Q1/Q2 of 2024....


That's a bleak future.

From
96C/192T Genoa vs 56C/112T Sapphire Rapids in 2022
128C/256T Bergamo vs 64C/128T Emerald Rapids in 2023

Bergamo and Emerald Rapids are two very different products targetting two very different markets. For the most part EMR is still competing with Genoa, and then GNR should compete against Turin.

 

[uzzi38]

"Apple is expected to release the majority of its devices with 3nm chips fabricated by TSMC in 2023"

TSMC on Track to Start 3nm Chip Production in Second Half of This Year

Apple chipmaking partner TSMC says it will be ready to move its 3nm chip process to volume production in the second half of this year, putting it on...

www.macrumors.com

You can bet AMD will not have N3 chips ready by 2023

Correction: you can bet that AMD actively decided against using N3 after the mess it is.

Jfc I've been ragging on about this for a while now for a reason.

 

[JasonLD]

Correction: you can bet that AMD actively decided against using N3 after the mess it is.

Not that it would matter much to AMD since they are not the first ones jumping to TSMC's latest nodes. Initial problems for TSMC's whatever latest node probably won't apply to AMD.

 

[nicalandia]

Bergamo and Emerald Rapids are two very different products targetting two very different markets. For the most part EMR is still competing with Genoa, and then GNR should compete against Turin.

You are using the word "Competing" very liberally.

 

[uzzi38]

Not that it would matter much to AMD since they are not the first ones jumping to TSMC's latest nodes. Initial problems for TSMC's whatever latest node probably won't apply to AMD.

Assuming their timelines wouldn't be thrown off just because they were never going to be the first to using the node is a terrible idea.

 

[uzzi38]

You are using the word "Competing" very liberally.

Hmmm... nope, I don't think I am.

 

[IntelUser2000]

I'm pretty sure that when they list a number of sockets, it's an "up to" number. 4S/8S being harder to support than 2S.

It's not. The roadmap has "4S/8S" and 2S.

 

[IntelUser2000]

Nope. Not bleak at all. It makes Intel look bad but AMD can't provide enough chips for all enterprise customers so Intel still sells a lot of server chips despite having fewer cores. Their customers will just buy double for double the amount of cores. What else can they do if AMD can't meet demand?

Also Arrowlake-S in possibly March/April of 2024 is not bleak at all. If they kept Meteorlake-S, it would have been that in September maybe.

If it's doable, then by all means they should skip Meteorlake-S for high-end.

 

[JasonLD]

Also Arrowlake-S in possibly March/April of 2024 is not bleak at all.

If Arrowlake-S is using either Intel 4 or 3, I would say that is possible but pretty much zero chance if Arrow Lake is using 20A.

 

[Exist50]

It's not. The roadmap has "4S/8S" and 2S.

How is that a contradiction? I'm saying that "2S" should be read as "only 1-2S", while "4S/8S" means "supporting the full 4S/8S". The 1-2S market is the biggest, so it would make no sense to launch with a platform that could only support a minority of the customer base. It's not technically easier either.

 

[IntelUser2000]

How is that a contradiction? I'm saying that "2S" should be read as "only 1-2S", while "4S/8S" means "supporting the full 4S/8S". The 1-2S market is the biggest, so it would make no sense to launch with a platform that could only support a minority of the customer base. It's not technically easier either.

Ok I get you now.

 

[nicalandia]

From:

Vision 2022 - 1 | Performance Index

"1-node, 2x AMD EPYC 7V73X on Azure HBv3, 128 cores (120 available), HT Off, Total Memory 448 GB OpenFOAM® v1912, Motorbike 28M @ 250 iterations Test by Intel as of 01/26/2022.  

1-node, 2x 4th Gen Intel Xeon Scalable processor (codenamed Sapphire Rapids, > 40 cores), HT On, Turbo On, Total Memory 512 GB OpenFOAM® v1912, Motorbike 28M @ 250 iterations; Build notes: Tools: Intel Parallel Studio Test by Intel as of 01/26/2022.


So let me get this straight A 112C/224T Sapphire Rapids is 2.1X faster than a gimped 120C/120T Milan-X? On the absolute best possible benchmark(Intel compiler and all)? Tough times ahead for Sapphire Rapids

 

[IntelUser2000]

So let me get this straight A 112C/224T Sapphire Rapids is 2.1X faster than a gimped 120C/120T Milan-X? On the absolute best possible benchmark(Intel compiler and all)? Tough times ahead for Sapphire Rapids

Actually on some very demanding HPC applications Hyperthreading is turned off because it performs better. Linpack for example performs higher when HT is off since Linpack saturates the FPU and the second thread has absolutely no room for enhancing performance.

SMT benefits performance by filling pipeline bubbles AND covering LLC misses. HPC applications LOVE memory bandwidth, the dataset is large, and fully saturates the execution units. Plus, even if HT was faster, it would have to clock lower because full AVX execution is very demanding anyway.