(8-17-13) AMD 9970 28nm vs. 20nm Poll

[RussianSensation]

The 7990 doesnt have frame pacing, only a driver that still doesnt work as well as NV hardware frame pacing...thats what I believe the respin is all about....

Jesus, say anything not in line with the knob ends in here and you are called out.....?

People didn't ignore your posts. My point if you release a card with a flawless driver in the form of HD7850x2 or 7870x2, no one would pay $550 for it. The rumors say that HD9970 is supposed to come in under $600. What does that say, presumably between $550 and $600. Even with hardware frame pacing, many PC gamers prefer a single fast GPU because they do not need to wait for SLI/CF profiles before their games run fast.

Some people do not want to buy a dual-GPU card and then have to wait for SLI/CF profiles.

Here is something else to think about. If HD9000 line has hardware frame pacing, then why wouldn't someone buy cheaper HD9850s-9870s in CF than a dual-GPU HD9970? The other point is if AMD decided to go with 2 mid-range chips to form a flagship card, they are sacrificing double precision performance. Well in that case, why not just take 2x lean HD7870 chips, and fuse them into one 425mm2 chip in the first place? Many of us believe AMD will not do that because they want to retain the full DP capabilities for HPC and their FirePro lines. This is why HD9970 is likely to be penalized again in performance / watt / mm2 compared to having a hypothetical 2x 7870 larger die chip.

Also, not a single rumor site has ever ran a story that HD9970 will be a dual-GPU card. It appears keysplayer started this rumor and people ran with it. I am not saying that AMD will release a 480-490mm2 9970 card but I can definitely see a 420-430mm2 card. Not sure why it's so impossible when NV has done plenty of 520-560mm2 chips in the past.

 

[SolMiester]

Guess we will have to wait and see then, no need for pointless graphs and walls of opinions mate.

 

[RussianSensation]

Guess we will have to wait and see then, no need for pointless graphs and walls of opinions mate.

It's not a pointless graph. It highlights major limitations of SLI/CF. You go out and spend $50+ on a new game and then it takes a while before your GTX690/7990 can get that needed 70-80% performance boost. How is that not relevant to the discussion when you yourself proposed the idea that AMD's next gen flagship card will be dual-GPU? I highlighted at least 3 reasons why this makes no sense for AMD: (1) price/performance would be really bad against just getting HD9850/9870 CF (2) Compute performance would suffer, which means might as well make a fused HD7870X2 to begin with (3) Since HD7990 already eclipses GTX780 and you can buy 7970 1Ghz/GE for near the price of a single 780, it makes little sense for AMD to do something like HD7850X2 as HD9970.

You also didn't comment at all that not a single credible source ever reported on 9970 being a dual-GPU card. Not NordicHardware, not Swecklockers, no one.

 

[Black Octagon]

Even with hardware frame pacing, many PC gamers prefer a single fast GPU because they do not need to wait for SLI/CF profiles before their games run fast.

It's not just scaling. Even the 690, which appears to be the best dual-GPU solution in terms of stuttering/frametimes, is behind single GPU solutions.

But in any case, the point is indeed that no-one has provided any credible reason for us to believe that the 9970 will be dual-GPU

 

[buletaja]

From Google Cache

Jason Tyler Scott
Member of Technical Staff - Fault Isolation Product Manager; Console/Server

• Electrical Fault Isolation - Device Analysis Department
• Responsible for electrical fault isolation on AMD's advanced 32nm, 28nm, 20nm APU / CPU architectures throughout; NPI, Qualification, Yield enhancement and Customer Return cycles
• Product owner for Game Console (Xbox / Playstation) and Server products

Steve Lacey
Contract Senior Mask Layout Designer at AMD

Contract Senior Mask Layout Designer
AMD
May 2013 – Present (4 months)Fort Collins, Co.
Provide layout support for the L2 and L3 Cache array group.
QualcommContract Mask Layout Designer IV

Qualcomm
November 2012 – May 2013 (7 months)Chandler, Az.
Analog/Mixed signal designs utilizing Cadence Virtuoso 6.1.5 in 20nm and 28nm technology.

Muhammad Anser

Senior Member of Technical Staff, 2010-2012
-----------------------------------------------------------------
As foundry interface and technology node owner, managed technology assessment and performance / power / cost trade-off determination to select optimal technology for dGPU and APU products. Developed and maintained infrastructure and methodologies to improve technology characterization and selection process.
* Managed foundry interface for TSMC 28nm and 20nm technology, resulting in 3 dGPU and 1 APU products being manufactured in 28nm technology.
* Obtained desired technology characteristics for high performance low power products by balancing design team requirements with foundry offering.
* Achieved higher Vmax for products by assessing and negotiating product reliability based on predicted product layout, thereby increasing product performance.
* Used design critical parameters list to negotiate technology characteristics with foundry, creating optimal device suite and reducing final product cost.

Russell Schreiber
SMTS Design Engineer at Advanced Micro Devices

July 2004 – Present (8 years 11 months)

Designed custom SRAM circuits for many of AMD's products and testchips from 65nm to current 20nm designs.

Seng Oon Toh

March 2011 – August 2013 (2 years 6 months)Sunnyvale, CA
Low-power advanced development of CPU and GPU.
Process technology optimization for 20nm Accelerated Processing Units (APU).
Lead EDA vendor interaction for enabling physical design in advanced technology nodes. Drive-to-yield test vehicle development.

Yuansheng Ma
engineer at Mentor Graphics

AMD/GLOBALFOUNDRIES
Privately Held; 10,001+ employees; Semiconductors industry
May 2006 – October 2012 (6 years 6 months)

Design Technology Co-optimization for 20nm, 14nm and 10nm,Lithography

Russell McMullan

20nm Technology Node Owner
AMD
2013 – Present (less than a year)

Allison Kuendiger
Senior Analog Engineer at AMD

March 2013 – Present (6 months)Markham, ON
*GF20LPM standard cell library development
--Schematic and layout optimization for digital logic cells on 20nm process.

Daniel Luke
Sr. Mask Layout Designer at AMD

Sr. Mask Layout Designer
AMD
2010 – 2012 (2 years) Austin, Texas Area

Leopard 20nm Project:Current work: Feasability Studies for L1 Sram Macros. Buildup of 128x2 columns and associated Datapath cells. First columns drawn were to a 28nm process, then redrawn in 20nm.

and much much more
Interestingly some of it mention GF20LPM

So 20nm this year or early next year still possibility

 

[raghu78]

Glo Fo is a joke. At 28nm they have been badly lagging TSMC. Only recently did volume production of some ARM SOCs for Rockchip begin. those too were on 28nm SLP and not 28nm high performance process.

http://www.globalfoundries.com/newsroom/2013/20130617.aspx

Even at 20nm TSMC is going to be far ahead of the others. GF has no credibility until they execute and have products out.

 

[buletaja]

Glo Fo is a joke. At 28nm they have been badly lagging TSMC. Only recently did volume production of some ARM SOCs for Rockchip begin. those too were on 28nm SLP and not 28nm high performance process.

http://www.globalfoundries.com/newsroom/2013/20130617.aspx

Even at 20nm TSMC is going to be far ahead of the others. GF has no credibility until they execute and have products out.

Fab 12 of TSMC is already 20nm full production, since March-April 2013 ( I will find the link, i reading somewhere)
Fab 14 & fab 15 is early next year

 

[raghu78]

Fab 12 of TSMC is already 20nm full production, since March-April 2013 ( I will find the link, i reading somewhere) Fab 14 & fab 15 is early next year

20nm is in risk production currently which means they are nowhere close to ready for commercial production of products for customers. for all you know yields could be practically 0 - 5%. 20nm has added complexity and cost due to double patterning. Other challenges include layout-dependent effects, new routing or local interconnect layers and more stricter design rules volume production begins in Q2 2014 but even that looks too optimistic now.

http://www.cadence.com/rl/Resources/white_papers/custom_20nm_wp.pdf
http://semimd.com/blog/2012/03/21/top-five-design-and-manufacturing-challenges-at-20nm/

here are some interesting comments to give a clue that 20nm is not going to significant next year.

TSMC Q1 2013 earnings call transcript

http://www.tsmc.com/uploadfile/ir/quarterly/2013/1J5NC/C/TSMC 1Q13 Transcript.pdf

read page 11

Andrew Lu - Barclays Capital - Analyst
I have two questions. Last time you mentioned 20 nanometer in next year revenue will be larger than 28 last year. How about the percentage in each quarter, which means the total percentage revenue will be higher compared to year 2012? For example, last year first quarter percentage 28 is about 5% but 22% by Q4.

Morris Chang - Taiwan Semiconductor Manufacturing Company Ltd -
Chairman and CEO
I'm not going to go into those details, but I repeat, I reiterate, I stand by what I said last time, that is that the volume of, production volume upward of 20 nanometers next year will be greater than 28 in 2012. That I stand by. But as to this percentage and so on

page 17 and 18

Steven Pelayo - HSBC - Analyst
Alright then. If you could just remind me what -- when is 20-nanometer revenues going to start being 1% or 2% or 3% of revenue? I think I forgot if you guys had commented.

Morris Chang - Taiwan Semiconductor Manufacturing Company Ltd - Chairman and CEO
20 nanometer?

Steven Pelayo - HSBC - Analyst
Yes.

Elizabeth Sun - Taiwan Semiconductor Manufacturing Company Ltd - Director of Corporate Communications
20.

Morris Chang - Taiwan Semiconductor Manufacturing Company Ltd - Chairman and CEO
When does it start?

Elizabeth Sun - Taiwan Semiconductor Manufacturing Company Ltd - Director of Corporate Communications
20 has not started

Steven Pelayo - HSBC - Analyst
I know. When do you expect?

Elizabeth Sun - Taiwan Semiconductor Manufacturing Company Ltd - Director of Corporate Communications
When, first half next year.

Lora Ho- Taiwan Semiconductor Manufacturing Company Ltd -SVP and CFO
You're asking about 20?

Morris Chang - Taiwan Semiconductor Manufacturing Company Ltd - Chairman and CEO
She answered already. She said first half of next year

Steven Pelayo - HSBC - Analyst
Ignore my first question. She's talking my second one.

Morris Chang - Taiwan Semiconductor Manufacturing Company Ltd - Chairman and CEO
You actually said a pretty low hurdle. You said 2% or 3%, right?

Steven Pelayo - HSBC - Analyst
The first few percent of 20 nanometer. What quarter will be the first few percentage of revenues will come from 20 nanometer for TSMC?

Morris Chang - Taiwan Semiconductor Manufacturing Company Ltd - Chairman and CEO
What quarter will be the 2% quarter?

Lora Ho- Taiwan Semiconductor Manufacturing Company Ltd- SVP and CFO
Based on our current estimation it will be roughly second quarter 2014

TSMC Q2 2013 earnings call transcript

http://www.tsmc.com/uploadfile/ir/quarterly/2013/2TGw1/E/TSMC 2Q13 Transcript.pdf

page 9 and 10

Michael Chou - Deutsche Bank - Analyst
Chairman, you gave guidance before regarding 20-nanometer sales portion in Q2 with low single digit. Do you have any revision for that or do you maintain the same guidance?

Morris Chang- Taiwan Semiconductor Manufacturing Co Ltd- Chairman, CEO
So Q2?

Elizabeth Sun - Taiwan Semiconductor Manufacturing Co Ltd - Director of Corporate Communication Division
Q2 next year, how much 20 nanometer will account for our revenue?

Morris Chang- Taiwan Semiconductor Manufacturing Co Ltd -Chairman, CEO
Well, Q2 next year, I don't know. But for whole year next year, I expect it will be the high single-digit percent of our total revenue.

Michael Chou - Deutsche Bank - Analyst
Thank you. Second question --.

Morris Chang- Taiwan Semiconductor Manufacturing Co Ltd -Chairman, CEO
I think that's correct, isn't it? For the whole year next year, I think 20 nanometers will be in the high single-digit percent of our total revenue.

page 12 and 13

Steven Pelayo - HSBC - Analyst
Okay. And then just one final question. This is really just a clarification. I think last quarter you spoke about 20 nanometer in its first year being bigger than what 28 nanometer was. This quarter you're now saying 20 nanometer is going to be high single-digit percentages of revenue, so I guess it depends what we think the total revenue number's going to be. I'm just curious, is there a change there on what you think the total dollar contribution can be next year for 28 nanometer -- for 20 nanometer at TSMC?

Morris Chang -Taiwan Semiconductor Manufacturing Co Ltd -Chairman, CEO
Well actually, the first year of 28 nanometer was 2011.

Steven Pelayo - HSBC - Analyst
I'm sorry.

Morris Chang- Taiwan Semiconductor Manufacturing Co Ltd -Chairman, CEO
The first year of 28 nanometer was 2011. And I'm quite sure that in 2011 28 nanometer did not reach a percent of our total revenue.

________________________________________________________
actually in seekingalpha the statement is

"The first year of 28-nanometer was 2011, and I'm quite sure that, in 2011, 28-nanometer did not reach 8% of our total revenue, yes."

http://www.xbitlabs.com/news/other/...Nearly_One_Third_of_TSMC_s_Wafer_Revenue.html

28nm wafer volume %
Q1 2012 - 5%, Q2 2012 - 7%, Q3 2012 - 13%, Q4 2012 - 22%

(5 + 7 + 13 + 22) / 4 = 11.75%

so there you have it. TSMC trying to evade and slip from their previous statements. TSMC is talking about dollar volume and not even wafer volume. we know 20nm wafers are very expensive. even more than 28nm at the start of its lifecycle due to double patterning. its looking very clearly now that 20nm is a H2 2014 event and the 20nm volume for 2014 is not going to be significant .

 

[buletaja]

Yes but, TSMC said FAB 12 is full production but it is no where > 40000 wafers/month

so basically in depends on your view

Fab 12 only probably wont enough to full fill the order

but yes Fab 12 already in full production it is not from random overclocking forum
or random pc forum, it is on eetimes and others and tsmc own site


http://www.eetimes.com/document.asp?doc_id=1280701

TSMC's phase 6 at Fab 12 in Hsinchu was slated to be TSMC's first 20-nm production site, coming on stream in 2013.

At a groundbreaking ceremony for phase 5 of Fab 14 held in April 2012 TSMC indicated that the fab would be TSMC's second 20nm-capable fab area and planned to begin 20-nm volume production early 2014.

http://www.simmtester.com/page/news/shownews.asp?num=15160

Following the equipment move-in, TSMC is expected to tape out SoC products at 20nm around the end of the second quarter with initial capacity of 5,000 12-inch wafer starts per month, the report cited unnamed fab tool suppliers as indicating. The new technology node is set to enter volume production in the third quarter with monthly capacity reaching more than 10,000 wafer starts, the report said.

TSMC in April 2012 disclosed that its 20nm technology would begin volume production at Phase 6 of its Fab 12 wafer fab (Hsinchu, northern Taiwan) in 2013, and Phase 5 of Fab 14 (Tainan, southern Taiwan) will be the foundry's second 20nm-capable fab, which is scheduled to enter volume production in early 2014.

TSMC own site
remember they installing 20nm on Fab 12
we already know that it is needed 6-12 month for full production

so it is so illogical if they installed on April 2012 with huge ceremony
and they still not passed risk production 1 year later

http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&newsid=6641

Tainan, Taiwan, R.O.C. – April 9, 2012 – TSMC (TWSE: 2330, NYSE: TSM) today held a groundbreaking ceremony at the South Taiwan Science Park in Tainan for Phase 5 of its Fab 14 GigaFab. Following TSMC's Fab 12 Phase 6 in Hsinchu, this facility will be another key production center for TSMC’s advanced 20nm technology, providing TSMC with abundant leading-edge capacity.

FAB 12

The Company's 20nm technology is scheduled to begin volume production at TSMC's Fab 12 next year.

What i am trying to said, in relation with 20nm TSMC:
20nm 2013 is already running, but the capacity wont near enough for >40000 wafers/month

There is many Gigafab at TSMC, but in general of course the output of 20nm production capacity is still low
Only 1 Gigafab ready for vol production for 2013.

 

[raghu78]

TSMC own site remember they installing 20nm on Fab 12 we already know that it is needed 6-12 month for full production

so it is so illogical if they installed on April 2012 with huge ceremony
and they still not passed risk production 1 year later

http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&newsid=6641

FAB 12

What i am trying to said, in relation with 20nm TSMC:
20nm 2013 is already running, but the capacity wont near enough for >40000 wafers/month

There is many Gigafab at TSMC, but in general of course the output of 20nm production capacity is still low Only 1 Gigafab ready for vol production for 2013.

What can I say ? thats an article about groundbreaking ceremony for expansion of Fab 14. you know it takes atleast 2 - 3 years to go from groundbreaking to completing construction, moving fab equipment and finally entering volume production.

 

[buletaja]

because Fab 12 is ready i am not talking FAB 14/15,

TSMC 2012
"The Company's 20nm technology is scheduled to begin volume production at TSMC's Fab 12 next year."

EEtimes
"monthly capacity reaching more than 10,000 wafer starts, the report said."

"TSMC's phase 6 at Fab 12 in Hsinchu was slated to be TSMC's first 20-nm production site, coming on stream in 2013."


and for Fab 14/15
"At a groundbreaking ceremony for phase 5 of Fab 14 held in April 2012 TSMC indicated that the fab would be TSMC's second 20nm-capable fab area and planned to begin 20-nm volume production early 2014."

 

[AtenRa]

No way it will be 20nm it will be 28nm.

 

[Gikaseixas]

conclusion: all guesses right now but the more likely one is 28nm

 

[Black Octagon]

If it's 20nm I'll eat my hat

(Disclaimer: I'm not wearing, nor do I own, a hat, and not because I ate the last one I owned)

 

[Smoblikat]

Its 28nm guys

http://forums.anandtech.com/showthread.php?p=35407248#post35407248

Hi guys, thank you for the welcome

It's 28

 

[BallaTheFeared]

Wonderful if true, my wallet can remain closed for another six+ months.

 

[3DVagabond]

Wonderful if true, my wallet can remain closed for another six+ months.

You never really believed they would release something that beat your 7950's for <$300?

 

[BallaTheFeared]

Oh no not that, just not enough to entice the switch over.

Bit different than last time, I was paying less than $150 per card on 40nm. This time I already sunk $600 so the chances I'll see a product for $600 or less that gets within 15% of what I've got are fairly good. It's just not going to happen on 28nm imo.

 

[buletaja]

Its 28nm guys

http://forums.anandtech.com/showthread.php?p=35407248#post35407248

he response to the CPU ---> Steamroller
which we already 99.99999% sure it is 28nm

And this thread is about VI, not Steamroller

 

[Phynaz]

because Fab 12 is ready i am not talking FAB 14/15,

TSMC 2012
"The Company's 20nm technology is scheduled to begin volume production at TSMC's Fab 12 next year."

EEtimes
"monthly capacity reaching more than 10,000 wafer starts, the report said."

"TSMC's phase 6 at Fab 12 in Hsinchu was slated to be TSMC's first 20-nm production site, coming on stream in 2013."


and for Fab 14/15
"At a groundbreaking ceremony for phase 5 of Fab 14 held in April 2012 TSMC indicated that the fab would be TSMC's second 20nm-capable fab area and planned to begin 20-nm volume production early 2014."

What part of "is scheduled" is it that you aren't getting?

Do you have a current link to a statement from TSMC that says they are currently in full 20nm production at FAB 12?

 

[Smoblikat]

he response to the CPU ---> Steamroller
which we already 99.99999% sure it is 28nm

And this thread is about VI, not Steamroller

Uh.......he posted in that in the video cards forum, I am 99.99999999999% sure VI is a video card.

 

[buletaja]

Uh.......he posted in that in the video cards forum, I am 99.99999999999% sure VI is a video card.

hmm on my opened thread
one post before he answer, is a question about Steamroller

Then we got this

Questions about CPUs? Ask Sam in CPUs. This is the Video forum.
-- stahlhart
_____________

Strange are the thread is changed ?

 

[videoclone]

Its a Titan killer but most likely on 28nm we will know more in a few months when its released. Prices are dropping on HD7970/50 all over the place to clear stock.
http://hardforum.com/showthread.php?t=1775645

 

[seitur]

More interesting question are:

1. When 20nm desktop GPUs will come? Is 2014 Q4 possible?

2. Will NV or AMD be first to deliver 20nm?

3. Prices of HD 9xxx series.

 

[raghu78]

More interesting question are:

1. When 20nm desktop GPUs will come? Is 2014 Q4 possible?

20nm in early Q3 2014 is looking the most realistic now.

2. Will NV or AMD be first to deliver 20nm?

both will be fairly close with their first 20nm chip.

3. Prices of HD 9xxx series.

Hawaii XT aka HD 9970 - USD 600
Hawaii Pro aka HD 9950 - USD 450 - 500
HD 9930 - USD 350 - 400

the HD 9800 chips are not yet revealed. but if they are they should take the USD 200 - 300 price range.