Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

There are other reasons for a product to be delayed than yields...

Such as?

And despite all this, it might just be cheaper for Intel to use a bunch of partically-defective tiles to build out a small volume of HEDT products. No new masks, etc. They're probably burning up a lot of 10ESF/Intel 7 wafers on Sapphire Rapids as it is. Why burn even more on a separate monolithic CPU?

I guess that MCC "24 Core Tile" was prepared for Birch Stream-AP LGA7529, DDR5-16ch, up-to-96cores and zen4-epyc competitor.

But probably now, it replaced by Emerald rapids-AP or Granite.

Links?

 

[ashFTW]

Such as?

Bugs for one. SPR has had a lot of steppings.

And despite all this, it might just be cheaper for Intel to use a bunch of partically-defective tiles to build out a small volume of HEDT products. No new masks, etc. They're probably burning up a lot of 10ESF/Intel 7 wafers on Sapphire Rapids as it is. Why burn even more on a separate monolithic CPU?

They will be making monolithic to satisfy large volume of low end servers. And repurpose some of these for HEDT

 

[DrMrLordX]

They will be making monolithic to satisfy large volume of low end servers. And repurpose some of these for HEDT

Guess we'll see for sure if they release them then. As for Sapphire Rapids having THAT many bugs, uh, okay? The delays have been long either way.

 

[Exist50]

Guess we'll see for sure if they release them then. As for Sapphire Rapids having THAT many bugs, uh, okay? The delays have been long either way.

Turns out that firing the entire server validation team has consequences. Another wonderful part of the BK legacy.

 

[lightisgood]

Bugs for one. SPR has had a lot of steppings.

You can say that agein.

Actually, SPRs is not full tile arch.
UPI, PCIe, accelerators,etc are not dissagregated.
SPRs simply breaks the limit of die-area-size by reticle.

So that, Gen-SPRs enables not only "1600mm^2 logical monolithic die"(LGA4677) but also "3000mm^2 logical monolithic die"(LGA7529).
But this is more complicate product than PVC.

 

[RTX]

AMD, since it has a separate IO die that contains all the memory channels and PCIe, doesn’t always have to use all 8 chiplets in every SKU. In non-stacked Milan parts, L3 can vary from 64MB to 256MB, which hints at fewer chiplets being used at times. For example the 32/64 7513 part has 128MB L3. This part can be made with 4 fully functional chiplets, or with 5 to 8 chiplets with varying numbers of core and L3 slices disabled. This gave AMD (starting with EPYC2) a lot of freedom, and it was a very frugal and brilliant design given their financial constraints.

Is this 4x8 cores with 4 dummy dies ( defects that couldn't be resold as Ryzen ) for the memory channels? It only has 4x32MB L3.

 

[ashFTW]

Is this 4x8 cores with 4 dummy dies ( defects that couldn't be resold as Ryzen ) for the memory channels? It only has 4x32MB L3.

The 4 dummy die could just be structural silicon, if need be. The socket could be stable with just the 4 functioning chiplets plus the IOD.

 

[DrMrLordX]

Turns out that firing the entire server validation team has consequences. Another wonderful part of the BK legacy.

Wait, what? That flew under my radar.

 

[Hotrod2go]

Raptor lake 'rumoured' to have DDR4/DDR5 memory controller?
Very curious to know for next upgrade. Videocardz has info on this but just speculation.
Any other sources on this aspect?

 

[IntelUser2000]

Raptor lake 'rumoured' to have DDR4/DDR5 memory controller?
Very curious to know for next upgrade. Videocardz has info on this but just speculation.
Any other sources on this aspect?

Searching shows it was leaked by a Chinese insider.

Raptorlake is October. So barely earlier than Alderlake. And that's for the Z790, and they are saying non K chips are coming in Q1.

Forget about Meteorlake P/U/H and desktop in Q2. Again, 5W is possible in Q2 but rest will not launch until late Q3/Q4 or even later. Since desktop is of miniscule volume, it means realistically Meteorlake is Jan 2024.

I guess the silver lining is that they can speed up Arrowlake desktop launch. If socketed desktops are practically nonexistent, maybe they can do K Arrowlake in Q2 of 2024.

 

[mikk]

If you refer to Meteor Lake-M 2+8, they would surely scale it up to 15W, they can replace Alder Lake-U 9/15 2+8 with a much higher performing solution.

 

[IntelUser2000]

If you refer to Meteor Lake-M 2+8, they would surely scale it up to 15W, they can replace Alder Lake-U 9/15 2+8 with a much higher performing solution.

No on the 9W and 15W, since it exists already. For early Meteorlake launch the only way that's possible is if they are skipping 9W and 15W for Raptorlake. I don't see that as a high chance.

And when you give minimum of 12 months between chips because you cannot shorten it unlike desktops, you end up with Raptorlake mobile in Q1 2023 and Meteorlake mobile in Q1 2024.

That's why the -M part(and I mean the true M like Core M and Lakefield) is the only one that can come in Q2 of next year. I also hope Meteorlake results in a proper M part rather than half-assed effort of Core M(includes Lakefield).

The staggered launch part might be that Meteorlake focuses on mobile while Arrowlake focuses on high performance(and desktop). Optimistically that's a possibility of Arrowlake-S launching earlier.

 

[mikk]

No on the 9W and 15W, since it exists already. For early Meteorlake launch the only way that's possible is if they are skipping 9W and 15W for Raptorlake. I don't see that as a high chance.

And when you give minimum of 12 months between chips because you cannot shorten it unlike desktops, you end up with Raptorlake mobile in Q1 2023 and Meteorlake mobile in Q1 2024.

That's why the -M part(and I mean the true M like Core M and Lakefield) is the only one that can come in Q2 of next year. I also hope Meteorlake results in a proper M part rather than half-assed effort of Core M(includes Lakefield).

The staggered launch part might be that Meteorlake focuses on mobile while Arrowlake focuses on high performance(and desktop). Optimistically that's a possibility of Arrowlake-S launching earlier.

Remember Cometlake-U and Icelake-U, they sure can do it within less than 12 months. The U9 and U15 segment is not big with ADL-U, they are focusing on ADL-P. Expecting MTL-M to be 5W only is not realistic and there is no reason why they shouldn't scale it up to 9W and 15W.

Keep in mind Alder Lake-M 5-7W was supposed to have only 1+4 cores. With MTL-M 2+8 they have bigger scaling options.

 

[nicalandia]

Wait, what? That flew under my radar.

Intel plans layoffs in data center group

The layoffs appear to reflect a repositioning of the data center group, which has grown into Intel’s most vital business

www.oregonlive.com

Large layoffs at Intel happening now - SemiAccurate

www.semiaccurate.com

 

[Exist50]

Intel plans layoffs in data center group

The layoffs appear to reflect a repositioning of the data center group, which has grown into Intel’s most vital business

www.oregonlive.com

Large layoffs at Intel happening now - SemiAccurate

www.semiaccurate.com

No, no, the validation team was laid off under BK, some years prior to this. Not even sure they were in Oregon.

 

[nicalandia]

10 Degrees is a substantial reduction in temperature.

https://www.tomshardware.com/news/thermal-grizzlys-contact-frame-reduces-alder-lake-temps-by-10-degrees-celsius

 

[jpiniero]

U15 segment is not big with ADL-U, they are focusing on ADL-P.

Just looking at Dell's website that is definitely wrong. There's plenty of 15 W U models. Looks like Tiger Lake is still what Corps are buying at this point though.

Meteor Lake taking over UP4 for 13th Gen makes a ton of sense and the rest is Raptor Lake. It's low volume and Intel marketing is obsessed with Apple so fanless designs is what they will push.

 

[nicalandia]

Intel marketing is obsessed with Apple so fanless designs is what they will push.

Apple is Never going back to Intel.

 

[jpiniero]

Apple is Never going back to Intel.

Correct but that doesn't mean Intel marketing isn't obsessed with them.

 

[uzzi38]

Turns out that firing the entire server validation team has consequences. Another wonderful part of the BK legacy.

Hold up a second Intel did WHAT

 

[Exist50]

Hold up a second Intel did WHAT

That's how I heard it, at least. Sometime during the BK era. Not sure of specifics, as I don't know anyone personally involved, but nonetheless.

Would certainly explain a lot though. What stepping is Sapphire Rapids on now? E? F? G?

 

[uzzi38]

That's how I heard it, at least. Sometime during the BK era. Not sure of specifics, as I don't know anyone personally involved, but nonetheless.

Would certainly explain a lot though. What stepping is Sapphire Rapids on now? E? F? G?

One of those yeah. Iirc Ice Lake had a tonne of revisions as well, so you're probably onto something there

 

[moinmoin]

That's how I heard it, at least. Sometime during the BK era. Not sure of specifics, as I don't know anyone personally involved, but nonetheless.

I remember reading about validation having to move faster/cut corners in the Spectre/Meltdown thread.

Page 2 - Massive security hole in CPU's incoming?Official Meltdown/Spectre Discussion Thread

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

"It's late in 2013. Intel is frantic about losing the mobile CPU wars to ARM. Meetings with all the validation groups. Head honcho in charge of Validation says something to the effect of: "We need to move faster. Validation at Intel is taking much longer than it does for our competition. We need to do whatever we can to reduce those times... we can't live forever in the shadow of the early 90's FDIV bug, we need to move on. Our competition is moving much faster than we are" - I'm paraphrasing. Many of the engineers in the room could remember the FDIV bug and the ensuing problems caused for Intel 20 years prior. Many of us were aghast that someone highly placed would suggest we needed to cut corners in validation - that wasn't explicitly said, of course, but that was the implicit message. That meeting there in late 2013 signaled a sea change at Intel to many of us who were there. And it didn't seem like it was going to be a good kind of sea change. Some of us chose to get out while the getting was good. As someone who worked in an Intel Validation group for SOCs until mid-2014 or so I can tell you, yes, you will see more CPU bugs from Intel than you have in the past from the post-FDIV-bug era until recently."

https://danluu.com/cpu-bugs/

 

[mikk]

Just looking at Dell's website that is definitely wrong. There's plenty of 15 W U models. Looks like Tiger Lake is still what Corps are buying at this point though.

Meteor Lake taking over UP4 for 13th Gen makes a ton of sense and the rest is Raptor Lake. It's low volume and Intel marketing is obsessed with Apple so fanless designs is what they will push.

It isn't wrong, check out the market as a whole. First ADL-H, then ADL-P and the last is ADL-U when it comes to device acceptance. Nobody denies they won't/can't use MTL-M for fanless designs, I'm just saying we will see these chips for the U9 or U15 segment as well, it's a perfect replacement for the current 2+8 ADL-U. We don't even know if RPT mobile is launching new U9 and U15 models, keep this in mind when you argue about the minimum 12 months rule between two generations and keep in mind this rule didn't exist for Icelake-U and Cometlake-U.

Tigerlake UP4 has a TDP range of 7-15W on paper, in real OEMs can set it to whatever they want.

Although it is actually a more efficient Tiger-Lake UP4 chip, the TDP values are set generously with a long-term value of 20 watts and a short-term value of 40 watts.

Lenovo ThinkBook 13x G1 in review: Slim notebook with good battery life

We test the Lenovo ThinkBook 13x G1-20WJ001KGE with Intel Core i7-1160G7, Intel iris Xe Graphics G7 96EUs, 16 GB RAM and 512 GB SSD.

www.notebookcheck.net

This isn't 1+4 Lakefield/ADL-M with a very limiting performance range, limiting MTL-M 2+8 to only 5W makes no sense. OEMs want a range they can work with.

 

[jpiniero]

It isn't wrong, check out the market as a whole. First ADL-H, then ADL-P and the last is ADL-U when it comes to device acceptance.

If you are implying that you think that H is the biggest seller because Intel launched it first... yeah no.

keep this in mind when you argue about the minimum 12 months rule between two generations and keep in mind this rule didn't exist for Icelake-U and Cometlake-U.

That's because Comet and Ice were meant to coexist together as part of one generation.