SemiAccurate Forums  

 
Go Back   SemiAccurate Forums > Main Category > CPUs

CPUs Talk about processors and related technology


Reply
 
Thread Tools Display Modes
  #41  
Old 08-10-2016, 04:10 AM
pTmd pTmd is offline
2^10
 
Join Date: Nov 2010
Posts: 1,992
pTmd is on a distinguished road
Default

Quote:
Originally Posted by Dresdenboy View Post
OK, they might technically do it. But there are now 2 complexes in SR, disabling cores in just one of them might cause some imbalances. And disabling cores in both CCX' might create an unusal topology.
You can always fuse off things symmetrically among complexes, if the only thing being shared by the cores is the L3 cache (incl. the coherent bus interface if you assume they are going for an inclusive L3).

For things within the complex, it depends on the design. If you assume the L3 cache is banked like Jaguar's L2 with a full crossbar, it should be perfectly fine to fuse them off.

Last edited by pTmd; 08-10-2016 at 04:13 AM.
Reply With Quote
  #42  
Old 08-10-2016, 11:55 AM
sdlvx sdlvx is offline
2^11
 
Join Date: Nov 2011
Posts: 2,532
sdlvx will become famous soon enough
Default

Saw this on WCCF tech. Claims Samsung 10nm mobile chips will hit 4ghz.
http://wccftech.com/samsung-exynos-8...pecifications/

Are GloFo and Samsung going to be working together for 10nm like they did for 14nm? If Zen is not clocking as well as we like, it looks like 10nm might have some hope for Zen+.
__________________
Within 27% margin of error and 3mm off my estimate of something 4mm, good enough to predict the future
Reply With Quote
  #43  
Old 08-10-2016, 12:26 PM
JeeBee JeeBee is online now
2^11
 
Join Date: Jan 2010
Location: UK
Posts: 3,362
JeeBee will become famous soon enough
Default

I read they weren't working together on 10nm.

But that doesn't mean that GF's 10nm will to plan and a similar bail-out won't occur.

The 10nm clock jump reported is quite amazing/suspicious. Maybe 14nm wasn't all that for Samsung (and hence GF).
Reply With Quote
  #44  
Old 08-10-2016, 12:49 PM
JohnJay JohnJay is offline
8-bit overflow
 
Join Date: Jun 2010
Posts: 375
JohnJay will become famous soon enough
Default

I'd read that GF may skip 10nm entirely.
Quote:
Originally Posted by SemiWiki.com
They are working on 10nm but focused on 7nm. They haven't announced whether they will do 10nm. He did mention that some customers have told them they will skip 10nm for 7nm. My understanding is that 10nm is a kind of intermediate node mostly targeted at Cell Phones Applications Processors. That isn't a big segment for Global Foundries so personally I wouldn't be surprised if they skipped 10nm and went right to 7nm, but Gary wouldn't comment on this yet.
https://www.semiwiki.com/forum/conte...y-forward.html
Reply With Quote
  #45  
Old 08-10-2016, 04:15 PM
testbug00 testbug00 is offline
>intel 4004
 
Join Date: Dec 2012
Location: Benicia
Posts: 7,242
testbug00 will become famous soon enough
Default

Wonder how good IBM's design tools were before GloFo bought them? If an external company were to come in and want to use their processes.

"working together" might just mean some license of current/former IBM tech to Samsung.

Where Samsung pieces it together, and GloFo gets a "free" license to use it?
__________________
-Q
Reply With Quote
  #46  
Old 08-10-2016, 05:33 PM
Moral Hazard Moral Hazard is offline
2^10
 
Join Date: Aug 2009
Location: Basket of Deplorables, America
Posts: 1,291
Moral Hazard is on a distinguished road
Default

Quote:
Originally Posted by JohnJay View Post
I'd read that GF may skip 10nm entirely.

https://www.semiwiki.com/forum/conte...y-forward.html
The GlobalFoundries guys spent a good chunk of this article on 7nm talking about how terrible 10nm was. Not sure if they're doing 10nm but they seem to think it's a lot like 20nm in that only a few customers (like the cell phone makers) will see an advantage. I'll be interested to see who actually does make 10nm chips and when.

I think we've entered a period where Moore's Law (i.e. the number of transistors on the chip that has the minimum cost/transistor doubles every two years) has been replaced by "The name of the the leading process of the leading fabs will contain a length that, when squared, will be 1/2 the length of their lead process from two years prior."

Admittedly they might reach a point where they are too embarrassed by how inaccurate the number is, such as if they go below the width of most atoms (I believe carbon is about .15nm for instance) or it they claim a value shorter than the Planck length (~1.62x10^-35).
__________________
“Be polite, be professional, but have a plan to kill everybody you meet.” - Gen Mad Dog Mattis

"I disapprove of what you say, but will defend to the death your right to say it." - Voltaire
Reply With Quote
  #47  
Old 08-10-2016, 06:01 PM
testbug00 testbug00 is offline
>intel 4004
 
Join Date: Dec 2012
Location: Benicia
Posts: 7,242
testbug00 will become famous soon enough
Default

Meh. New 10nm process are just real 14nm process.

By talking about 7nm, GloFo might mean a real 10nm process.
__________________
-Q
Reply With Quote
  #48  
Old 08-10-2016, 07:44 PM
Moral Hazard Moral Hazard is offline
2^10
 
Join Date: Aug 2009
Location: Basket of Deplorables, America
Posts: 1,291
Moral Hazard is on a distinguished road
Default

Quote:
Originally Posted by testbug00 View Post
Meh. New 10nm process are just real 14nm process.

By talking about 7nm, GloFo might mean a real 10nm process.
Supposedly Intel has a true 14nm process. Nvidia put 15 billion transistors in a 600mm2 Tesla P100 chip on TSMC's 'really just 20nm' 16FF+ process while intel put 9 billion transistors in their 700mm2 Knights Landing chip on their 'it's really 14 nm honest' 14nm process. Of course one is a compute accelerator and the other is only sometimes a comput accelerator so....yeah.
__________________
“Be polite, be professional, but have a plan to kill everybody you meet.” - Gen Mad Dog Mattis

"I disapprove of what you say, but will defend to the death your right to say it." - Voltaire
Reply With Quote
  #49  
Old 08-10-2016, 08:17 PM
testbug00 testbug00 is offline
>intel 4004
 
Join Date: Dec 2012
Location: Benicia
Posts: 7,242
testbug00 will become famous soon enough
Default

Quote:
Originally Posted by Moral Hazard View Post
Supposedly Intel has a true 14nm process. Nvidia put 15 billion transistors in a 600mm2 Tesla P100 chip on TSMC's 'really just 20nm' 16FF+ process while intel put 9 billion transistors in their 700mm2 Knights Landing chip on their 'it's really 14 nm honest' 14nm process. Of course one is a compute accelerator and the other is only sometimes a comput accelerator so....yeah.
Yes, because a process designed with experienced refined over years for making CPUs translates well into making devices which are similar to GPUs.

Intel's process probably isn't truly 14nm, but relative to their 22nm process it is a full shrink.
__________________
-Q
Reply With Quote
  #50  
Old 08-10-2016, 08:31 PM
Moral Hazard Moral Hazard is offline
2^10
 
Join Date: Aug 2009
Location: Basket of Deplorables, America
Posts: 1,291
Moral Hazard is on a distinguished road
Default

I would try doing a more explicitly "Apples to Apples" comparison between Intel's 14nm process and TSMC's 16 and/or Samsung's 14nm process but I can't find a transistor count for any of the foundry based mobile chips (Qualcomm 820, Exynos 8, or Apple A9) and I know cell phone chips tend to be dense for space reasons and because they run at low frequency/voltage so even that wouldn't be the same.

When Zen comes out we should at least get some idea about how the density at GlobalFoundries compares (assuming they give out transistor counts which I suspect they will) and in turn that will tell us about Samsung since it's the same process.

Actually we should be able to tell something about TSMC once somebody gives the dimensions of the 'Xbox One 'S'' chip assuming it either has about the same 5 billion transistors as the original or they give us an updated number.

edit: Saw your post after I posted so I'm adding this
Quote:
Originally Posted by testbug00 View Post
Yes, because a process designed with experienced refined over years for making CPUs translates well into making devices which are similar to GPUs.

Intel's process probably isn't truly 14nm, but relative to their 22nm process it is a full shrink.
Well Intel does make integrated GPUs a large part of the dies on their current chips, and they do tend to perform pretty awfully given how much space they take up. It has been suggested (by people who know a lot more than me) that their 1D process is very good for making processors but not very flexible (and flexibility is an obvious requirement for fabs, possibly why they don't seem to get any fab business from companies they don't end up buying).
__________________
“Be polite, be professional, but have a plan to kill everybody you meet.” - Gen Mad Dog Mattis

"I disapprove of what you say, but will defend to the death your right to say it." - Voltaire

Last edited by Moral Hazard; 08-10-2016 at 08:39 PM.
Reply With Quote
Reply


Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


Forum Jump


All times are GMT -5. The time now is 03:50 AM.


Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2017, vBulletin Solutions, Inc.
SemiAccurate is a division of Stone Arch Networking Services, Inc. Copyright © 2009 Stone Arch Networking Services, Inc, all rights reserved.