Discussion Intel current and future Lakes & Rapids thread

Page 700 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Kocicak

Senior member
Jan 17, 2019
478
388
106
Is this the most steppings of one CPU in the history of mankind?

BTW how much does each new stepping cost? I believe the changes may be made just in a few layers of the CPU, so not all the reticles need to be updated?
 
Last edited:

Exist50

Golden Member
Aug 18, 2016
1,140
1,138
136
Is this the most steppings of one CPU in the history of mankind?

BTW how much does each new stepping cost? I believe the changes may be made just in a few layers of the CPU, so not all the reticles need to be updated?
IIRC, Broadwell made it to F, and if you go back far enough, there might be some higher numbers from when additional steppings were faster/cheaper. But in terms of money and time spent, SPR quite possibly holds the record.

Oh, and as to your question, the letter indicates the base/transistor layer and all the metal layers were changed, while the number indicates only metal layers (which ones/ how many unknown) were changed.
 

Exist50

Golden Member
Aug 18, 2016
1,140
1,138
136
but it's already accounted for on the design/build costs
Well, projects usually have a planned number of steppings that are budgeted in, but in the case of something like SPR, the company just has to eat the (enormous) extra cost. That, or give up.
 

DrMrLordX

Lifer
Apr 27, 2000
19,941
8,915
136
IIRC, Broadwell made it to F, and if you go back far enough, there might be some higher numbers from when additional steppings were faster/cheaper. But in terms of money and time spent, SPR quite possibly holds the record.
Did Broadwell make it to F before it even launched?
 

DrMrLordX

Lifer
Apr 27, 2000
19,941
8,915
136
Maybe? I found this article, but it seems a little unclear about what actually happened with the E-step.
Ohhhh you meant the client parts! Okay I can see that. Broadwell never made it to desktop (except for the -C parts) and took awhile to make it to mobile.
 
Last edited:

Kocicak

Senior member
Jan 17, 2019
478
388
106
Wow, they said that 14 nm die required 50 different masks. I wonder what is the number for todays processors.
 

moinmoin

Diamond Member
Jun 1, 2017
3,782
5,519
136
I believe that one layer of the CPU may require few differents masks.
If a mask is too coarse to create a layer multiple masks are used per layer indeed. That's what's called double or quad pattering (two and four masks respectively). The simplification of going from DUV to EUV is exactly by reducing the amount of masks by no longer using that multiple pattering for printing finer details. Wikipedia goes pretty deep into this topic for a change.
 

John Carmack

Member
Sep 10, 2016
143
197
116

nicalandia

Platinum Member
Jan 10, 2019
2,146
3,204
106
Did Broadwell make it to F before it even launched?
According to Dr. Ian Cutress Intel is past F.. And its currently on G0 and not ready to Ramp Up

Screenshot_20220919-071950_Chrome.jpg


A I am with Igor and Locuza here.

A0, A1, B0, C0, C1, C2, D0, E0, E2, E3,E4,E5. It would seem that they did not stay too much on F, so let say F0 was the only F stepping then Jumped to G0 if that is the case then Intel has gone to at least 14 Steppings on Sapphire Rapids.
 
Last edited:

Exist50

Golden Member
Aug 18, 2016
1,140
1,138
136
According to Dr. Ian Cutress Intel is past F.. And its currently on G0 and not ready to Ramp Up

View attachment 67776


A I am with Igor and Locuza here.

A0, A1, B0, C0, C1, C2, D0, E0, E2, E3,E4,E5. It would seem that they did not stay too much on F, so let say F0 was the only F stepping then Jumped to G0 if that is the case then Intel has gone to at least 14 Steppings on Sapphire Rapids.
I could certainly believe 14 steppings, but that ordering seems odd. It appears that they're shipping on E5, maybe E6? So why would they then do two more full steppings so soon after? Usually companies do post-launch steppings to improve yields or bin split, but then why two? I almost wonder if there's some confusion about counting. Intel often separates the stepping letters for different dies (e.g. A0 for the first die, P0 or something for the second), but if one runs far enough over, they could be forced to skip or do something else weird. Or maybe the G0 is the smaller die that's rumored? Will certainly be interesting to see what actually makes it to market.
 

Kocicak

Senior member
Jan 17, 2019
478
388
106
I dont know much about CPU development, but I thought that there are some simulation tools available to avoid redoing the silicone again and again.

Where does the information, that it is still not ready, come from? Would it be possible to get some intel about what is currently wrong with it?

Learning about the past problems and the process of fixing it would be most interesting to many I believe. Intel should publish some study about it to improve general knowledge of processor development.
 

Thala

Golden Member
Nov 12, 2014
1,335
645
136
I dont know much about CPU development, but I thought that there are some simulation tools available to avoid redoing the silicone again and again.
Sure, there is RTL and gate level simulation available for design verification. However the coverage on system level typically is very low for various reasons - one is simulation speed (10^6 times slower than real-time or even less).
 

igor_kavinski

Diamond Member
Jul 27, 2020
4,723
2,968
106
Found this, thanks to Phoronix: Brown-Neri LPC 2022.09.13 Linux Intel-Hybrid Scheduling

1663663214246.png

1663663275165.png

Running AVX2 code on E-core suffers more than running SSE.

If I was making a game engine, I would just blacklist E-cores and call it a day. The time saved could be spent doing something better.

Maybe now we know why AVX-512 isn't on E-cores. Not only would it be more power hungry, it wouldn't also be very performant.
 
Last edited:

igor_kavinski

Diamond Member
Jul 27, 2020
4,723
2,968
106
I dont know much about CPU development, but I thought that there are some simulation tools available to avoid redoing the silicone again and again.
The problem seems to be that such simulation tools for a new manufacturing process are constantly in flux, as new and new data comes in about the various characteristics of the new process.

With each new stepping, they are probably updating the simulation tools in tandem, to prevent the same bugs from appearing in future silicon.
 

Kocicak

Senior member
Jan 17, 2019
478
388
106
Perhaps the goal is not a finished processor, but the development process itself. To keep people employed and entertained.

That reminds me Hunger wall of Prague, a project believed to be built just to give work and food to the poor.
 
Last edited:

ASK THE COMMUNITY