Go Back   nV News Forums > Hardware Forums > Rumor Mill

Newegg Daily Deals

Closed Thread
 
Thread Tools
Old 05-01-11, 04:25 PM   #1
josiahsuarez
Registered User
 
josiahsuarez's Avatar
 
Join Date: Feb 2008
Location: Dallas, Texas
Posts: 739
Default nVidia's 28nm Kepler expected to hit stores in late 2011?

http://www.brightsideofnews.com/news...worldwide.aspx
Quote:
Personally, I expect that nVidia will align the GPU Technology Conference 2012 in San Francisco Bay Area with the official unveiling of Tesla boards based on Kepler, their next-gen 28nm architecture. If rumors are to believe, due to problems TSMC is experiencing with the 28nm process the initial 28nm silicon from AMD will be nothing more than a die-shrink of current 40nm GPUs (AMD Radeon HD 6000 Series). nVidia's 28nm Kepler though is expected to hit the stores as a consumer part in late 2011 and appear as a mission-critical part a few months afterwards. After all, as Teslas or Quadros, GPUs have no room for any silicon-related issues.


new Nvidia GPU late 2011?
josiahsuarez is offline  
Old 05-02-11, 02:04 AM   #2
jlippo
Registered User
 
Join Date: Mar 2006
Posts: 532
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by josiahsuarez View Post
It's about time as well, we haven't had anything new sience Fermi.
Fermi had around 3B transistors so I would expect Kepler to have around 5-6B transistors.
jlippo is offline  
Old 05-02-11, 05:10 AM   #3
LydianKnight
Registered Knight
 
LydianKnight's Avatar
 
Join Date: Jan 2009
Location: Málaga, Spain
Posts: 977
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Isn't that going a bit too far on the transistor count? I mean...

Kepler isn't the 'Tock' (in Intel terms), but more like the 'Tick', so... don't know, just it seems a bit too much for me (unless they're planning to introduce something like 680 (768c) / 670 (640c) / 660 (512c) or something like that...)

LydianKnight is offline  
Old 05-02-11, 06:20 AM   #4
jlippo
Registered User
 
Join Date: Mar 2006
Posts: 532
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by LydianKnight View Post
Isn't that going a bit too far on the transistor count? I mean...

Kepler isn't the 'Tock' (in Intel terms), but more like the 'Tick', so... don't know, just it seems a bit too much for me (unless they're planning to introduce something like 680 (768c) / 670 (640c) / 660 (512c) or something like that...)

They did double transistor count when moving from 8000 series to GTX280 and have pretty much doubled transistor count with every 'Tick' in their history.
jlippo is offline  
Old 05-02-11, 11:36 AM   #5
josiahsuarez
Registered User
 
josiahsuarez's Avatar
 
Join Date: Feb 2008
Location: Dallas, Texas
Posts: 739
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

don't be fooled by that graph; "DP GFLOPS per Watt"

the graph is about efficiency, not absolute performance.
josiahsuarez is offline  
Old 05-02-11, 10:35 PM   #6
shadow001
Registered User
 
Join Date: Jul 2003
Posts: 1,526
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by josiahsuarez View Post
don't be fooled by that graph; "DP GFLOPS per Watt"

the graph is about efficiency, not absolute performance.


True, but going from a 40 nm fab process, which is what fermi still uses as we all know, to a 28nm process is a 60% reduction(transistor lenght and height), so if the ratio between logic circuits and cache is similar between kepler and fermi(cache can be packed closer together), it allows a 60% increase to the overall transistor budget while maintaining the same die sizes and production costs(if yeilds are good of course), and fermi is packing 3.2 billion transistors.


So 3.2 billion transistors at 40nm + 60% and assuming the same die size for both fermi and kepler = 5.1 billion transistors at 28nm for kepler....Obviously i haven't considered power consumption or cooling issues here, wich are variables we'd only know if TSMC was willing to divulge said information about their 28nm process( unlikely without having a licence with TSMC...).


Fermi is rated at 240 watts TDP and that's only 60 watts short of the maximum limit for PCI-e compliance(300 watts), so can Kepler double performance over fermi with the 60% increase in transistor budget, cost the same to produce as fermi since it's the same die size, and remain under that 300 watt limit for PCI-e compliance?.... Wait 6 months to find out(if not more)...
shadow001 is offline  
Old 05-02-11, 10:51 PM   #7
shadow001
Registered User
 
Join Date: Jul 2003
Posts: 1,526
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Now that i really payed attention to the chart, it lists 5 gigaflop at dual precision floating point per watt, so it's not that hard to compare with fermi on a pretty rough basis....


let's say for sake of argument that Kepler uses the same power as fermi(it can't go much higher anyways because of PCI-e certification issues)....Keep in mind this is very rough math and in no way meant to be precise, just a rough estimate.



240 watts average x 5 GF per watt = 1.2 terraflop, dual precision floating point math for Kepler, but if the card uses closer to 300 watts, then it could be as much as 1.5 terraflop dual precision floating point math...


Fermi is rated at about 750 Giga flop for dual precision floating point in the GTX580 cards with all 512 shaders enabled at 800 Mhz clock, so kepler might actually be pretty close to doubling the shader power overall....No idea for texturing, or fillrate or tesselation though...
shadow001 is offline  
Old 05-03-11, 02:49 AM   #8
jlippo
Registered User
 
Join Date: Mar 2006
Posts: 532
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by shadow001 View Post
No idea for texturing, or fillrate or tesselation though...
They are adding more shaders and units, so it is pretty much guaranteed to have more polymorh engines and thus increasing tesselation speed.
I would love them to open little bit their setup engine for non-tesselated polygons, currently the full speed is reserved for Quadros.

Also if they do some nice little tweaks here and there, they can improve the general speed significantly as Fermi had some pitfalls.
jlippo is offline  

Old 05-03-11, 07:18 AM   #9
TheANIMAL
Registered User
 
TheANIMAL's Avatar
 
Join Date: Feb 2008
Location: Near Bude, Cornwall, UK
Posts: 1,291
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Hold on...

40 divided by 28 is 1.42857....

1.42857... squared is 2.0408... so you can get a double the number of transistors in the same space moving from 40 to 28nm. That's just over 6.4 billion transistors unless my maths is completely off.

The reduction has to be squared to make it work for a square chip doesn't it?
TheANIMAL is offline  
Old 05-03-11, 12:11 PM   #10
shadow001
Registered User
 
Join Date: Jul 2003
Posts: 1,526
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by TheANIMAL View Post
Hold on...

40 divided by 28 is 1.42857....

1.42857... squared is 2.0408... so you can get a double the number of transistors in the same space moving from 40 to 28nm. That's just over 6.4 billion transistors unless my maths is completely off.

The reduction has to be squared to make it work for a square chip doesn't it?

There's a 30% reduction in size going from 40nm to 28nm, but since we have to consider that both transistor lenght and the overall their diameter get affected with the change in fabrication process, that increases to 60% basically....


If the calculations for the transistors are made for all 3 dimensions( lenght, height and how wide they are), then the 30% reduction applies in all 3 dimensions, and the overall transistor count can increase by 90% and as you said, nearly double the transistor budget for the same die size compared to 40nm.....


It really comes down to the overall shape of the transistor and overall layout, and logic to cache ratios and power consumption and thermals, but it's easy to see that it'll allow a transistor budget increase that's significant to say the least, and will allow to enhance the GPU in many aspects over fermi, that much is sure....
shadow001 is offline  
Old 05-03-11, 01:49 PM   #11
TheANIMAL
Registered User
 
TheANIMAL's Avatar
 
Join Date: Feb 2008
Location: Near Bude, Cornwall, UK
Posts: 1,291
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by shadow001 View Post
There's a 30% reduction in size going from 40nm to 28nm, but since we have to consider that both transistor lenght and the overall their diameter get affected with the change in fabrication process, that increases to 60% basically....


If the calculations for the transistors are made for all 3 dimensions( lenght, height and how wide they are), then the 30% reduction applies in all 3 dimensions, and the overall transistor count can increase by 90% and as you said, nearly double the transistor budget for the same die size compared to 40nm.....


It really comes down to the overall shape of the transistor and overall layout, and logic to cache ratios and power consumption and thermals, but it's easy to see that it'll allow a transistor budget increase that's significant to say the least, and will allow to enhance the GPU in many aspects over fermi, that much is sure....
True that.

Using maths and mixing it with reality is usually where i fall down.
TheANIMAL is offline  
Old 05-03-11, 02:35 PM   #12
shadow001
Registered User
 
Join Date: Jul 2003
Posts: 1,526
Default Re: nVidia's 28nm Kepler expected to hit stores in late 2011?

Quote:
Originally Posted by TheANIMAL View Post
True that.

Using maths and mixing it with reality is usually where i fall down.

keep in mind that there's still other factors coming into it as well, such as the amount of room needed for the connections between each transistor, and how many vertical layers will be used as well(think of them as individual floors of a building), and the number of vias used as well(used to connect the individual layers vertically)...


The largest one of all is the gate portion of the transistor, wich allows the transistor to turn on or off and is the fundamental requirement to create a logic circuit between several transistors, and as the entire transistor gets smaller and smaller with each fabrication process reduction, the gate portion of that becomes absolutely tiny, likely not being more than a few atoms wide as the fab processes get smaller and smaller...


It's kinda hard to control the passage of electrical current when it's that narrow, as it might just pass right thru regardless of the gate in the middle of the transistor is open or closed, or what is usually referred to as leaking.....


Now add Radio frequency interference and electromagnetic interference testing to the mix, since the video card will be right next to other components within the PC, and should not cause interference with those parts in those 2 aspects, and the actual design of the architecture itself, and it becomes easy to understand why a new design takes 3~4 years to develop, and they have to do it for a specific fabrication process, a few years before that process is ready for mass production and when consumers can buy the end product...


It's a pretty big gamble, as hundreds of millions or even a few billion, might already have been spent developing it, before a single one is available for purchase by consumers...
shadow001 is offline  
Closed Thread


Thread Tools

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 08:41 PM.


Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright ©1998 - 2014, nV News.