Discussion Intel current and future Lakes & Rapids thread

Page 798 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hulk

Diamond Member
Oct 9, 1999
5,138
3,727
136
Perhaps Intel 4/MTL is going better than we think and the Intel 7/Raptor Refresh is going to comprise 14 generation higher end while MTL, which may well be cheaper to manufacture will take the mid to lower shelf parts?
 

bsp2020

Member
Dec 29, 2015
106
122
116
Perhaps Intel 4/MTL is going better than we think and the Intel 7/Raptor Refresh is going to comprise 14 generation higher end while MTL, which may well be cheaper to manufacture will take the mid to lower shelf parts?
Why do you think MTL may be cheaper to manufacture? Using tiles would minimize the increase but it is not going to make it cheaper. At least, not for a while if ever.
 

BorisTheBlade82

Senior member
May 1, 2020
709
1,131
136
Why do you think MTL may be cheaper to manufacture? Using tiles would minimize the increase but it is not going to make it cheaper. At least, not for a while if ever.
Well, maybe not in the short term. But in the long term it is exactly the way to go. Just look at the ridiculous die sizes of EMR - and it will bring them nowhere near to competitiveness. Even an RTL successor would not be sustainable on newer nodes with all its badly scaling bloat on the same die.
 

bsp2020

Member
Dec 29, 2015
106
122
116
Well, maybe not in the short term. But in the long term it is exactly the way to go. Just look at the ridiculous die sizes of EMR - and it will bring them nowhere near to competitiveness. Even an RTL successor would not be sustainable on newer nodes with all its badly scaling bloat on the same die.
Hopefully. But I think the economic part of Moore's law is dead. Performance and area scaling seems to be still happening. So unless you build a chip mixing and matching different process node, the cost savings won't be great even if you break up your giant chip into pieces.

That's the reason why I don't think Intel will catch up to AMD in the server market until at least Granite Rapids comes out. I don't know if Granite Rapids will use multiple process nodes. But Sapphire Rapids and Emerald Rapids does not stand a chance.

On the desktop side, Intel's choice of using silicon interposer will limit how fast they can bring the cost down. It will be interesting to see what AMD does when they introduce real chiplet based APU. I'm sure they will delay using silicon interposer for as long as possible.
 
  • Like
Reactions: Tlh97 and moinmoin

Hulk

Diamond Member
Oct 9, 1999
5,138
3,727
136
Why do you think MTL may be cheaper to manufacture? Using tiles would minimize the increase but it is not going to make it cheaper. At least, not for a while if ever.
The reason for chiplets is economics so that is why I'm thinking that. Intel may be looking at transitioning from monolithic so chiplets sooner rather than later. Besides increasing yields by ditching the monolithic designs they can optimize process for various parts of the design.
 

BorisTheBlade82

Senior member
May 1, 2020
709
1,131
136
@bsp2020
Agreed. I'd say that AMD leapfrogs interposers for InFO-R and/or EFB.
One aspect that you didn't mention is the reticle limit, making dies like EMR impossible at High-NA.
So both of them net to get themselves sorted out sooner rather than later.
 

A///

Diamond Member
Feb 24, 2017
4,351
3,160
136
The reason for chiplets is economics so that is why I'm thinking that. Intel may be looking at transitioning from monolithic so chiplets sooner rather than later. Besides increasing yields by ditching the monolithic designs they can optimize process for various parts of the design.
I think what the other user seems to be confused about is that it is cheaper..... for the manufacturer or in the case for amd for amd to pay for and tsmc doesn't waste much product that makes them look bad. Amd isn't going to give you a discount if it costs them less! they will keep the difference and charge you as much as they would have with a monolithic die.

intel going tiles with their own interconnect should be mighty interesting given the performance penalty it'll incur via latencies.
 

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
That is a bit odd. It probably won't have much effect on adoption rate though.
I think it would. No idea if any of the cloud vendors will bother with EMR, but they absolutely hate having to certify new platforms. Wonder what the motherboard difference is, if anything.
 

DrMrLordX

Lifer
Apr 27, 2000
22,902
12,971
136
I think it would. No idea if any of the cloud vendors will bother with EMR, but they absolutely hate having to certify new platforms. Wonder what the motherboard difference is, if anything.
Power delivery perhaps? Maybe there's a flaw in v1 of the platform? If Emerald Rapids is a substantially-better CPU then they'll do the validation one way or another. Sapphire Rapids can't be making vendors and system integrators very happy right now, except for those catering to the select audience that is better-served by Sapphire Rapids than competing products.
 

A///

Diamond Member
Feb 24, 2017
4,351
3,160
136
only those locked into contracts. the kill fee won't be a lot or it might but the validation costs and time to integrate amd now don't make a lot of sense for vendors or integrators or dc's.
 

Hulk

Diamond Member
Oct 9, 1999
5,138
3,727
136
Curious, what are the performance limits of 2 E-cores?
Idle?
Updating apps in the background?
Web browsing?
Watching youtube?
I think web browsing and YouTube would be fine. Most videos are off-loaded to the GPU anyway.

My old Broadwell is stuck at .79GHz for some reason but I can still browse and watch Netflix and such on it.
 
  • Like
Reactions: Tlh97 and Geddagod

BorisTheBlade82

Senior member
May 1, 2020
709
1,131
136
If I had to guess, I would say that web browsing will mostly take place on the compute tile for reasons of snappiness. Intel would not want to introduce stuttering - even if it was rather occasional.
The two SoC tile cores should really only be there for non time-critical background tasks like updates, cloud sync, AV, etc.

Also I would guess that these will be optimized for low voltage / low frequency.
 

Geddagod

Golden Member
Dec 28, 2021
1,524
1,620
106
Heavily suggests MTL will boost slightly upwards of 5 GHz if they are willing to release an i5 sku (the 13600k boosts up to 5.1 GHz) for ST.
 
  • Like
Reactions: Tlh97 and Exist50

Geddagod

Golden Member
Dec 28, 2021
1,524
1,620
106
Also ARL having lower TDP skus suggest MTL might end up being a stand alone generation rather than lumped in together with ARL, since ARL looks like it has its own low end skus.
 

Geddagod

Golden Member
Dec 28, 2021
1,524
1,620
106
there are I9 SKUs with lower TDP like the 13900 and 13900T
Good catch. Totally forgot about those haha.
The slide saying "up to" i9 makes me think that the i9 won't be the only sku, and we will se a full lineup, but maybe it's just the i9s and i7s then?
 

Henry swagger

Senior member
Feb 9, 2022
511
313
106
Heavily suggests MTL will boost slightly upwards of 5 GHz if they are willing to release an i5 sku (the 13600k boosts up to 5.1 GHz) for ST.
Raptor refresh must be good if the is no i9 for meteor lake
 

Hulk

Diamond Member
Oct 9, 1999
5,138
3,727
136
To me the real news of MTL will be that Intel 4 is up and running, yielding well and hitting decent clocks. Density and efficiency gains means Intel has a competitive process for the next architecture.
 

jpiniero

Lifer
Oct 1, 2010
16,810
7,253
136
To me the real news of MTL will be that Intel 4 is up and running, yielding well and hitting decent clocks. Density and efficiency gains means Intel has a competitive process for the next architecture.

Can't really say that, assuming that Arrow's tile is being fabbed at TSMC. The 6+8 MTL CPU tile is not that big and the rest of the product is being fabbed elsewhere.
 

AMDK11

Senior member
Jul 15, 2019
473
407
136
How can you be sure that ArrowLake-S is on TSMC's trial? So far I read that it is supposed to be on Intel 20A.