Sign in with
Sign up | Sign in
Your question

Poor Fermi Yields - Fudzilla Rumor's

Last response: in Graphics & Displays
Share
a b U Graphics card
January 12, 2010 4:59:10 PM

http://www.fudzilla.com/content/view/17205/1/

Claiming 20% fermi yields, (works out to $200 a GPU). Also claiming 20% faster than a 5870. Interesting none the less.
a b U Graphics card
January 12, 2010 5:00:40 PM

The ATI 5x00 yields are at 4%
If a person considers 20% to be poor, then 4% is worse than piss poor.
a b U Graphics card
January 12, 2010 5:09:09 PM

I think the 4% yields was a typo, even TSMC was saying 40% in December, I think 4% is not considered a yield, (about 2-5 chips per wafer). There is no way HD5870 was 4% because that would work out to about $500-1K per chip (not board, per GPU).

I think it's 20% vs 40%. which means it's still running at about 2:1 for chips cost between the two companies, just like during the GTX280/HD4K series.

a b U Graphics card
January 12, 2010 5:10:54 PM

Bluescreendeath said:
The ATI 5x00 yields are at 4%
If a person considers 20% to be poor, then 4% is worse than piss poor.


Where on earth does it say 4%? The article claims 60-80% for the 5870s... Even during the no 5870 to save a life months it was never claimed to be as low as 4%... below 40% was the worst I saw..
January 12, 2010 5:42:33 PM

The article has been taken down. The author made an update stating that 4% yield was in fact correct and took a lot of heat for saying so. 40% is what they were getting when there were fab issues for a month or so.

The heat was well deserved and the article really just seemed like some Nvidia fud spin. It's really sad that the article was even posted on this web site.
a b U Graphics card
January 12, 2010 6:13:07 PM

Lavacon said:
The article has been taken down. The author made an update stating that 4% yield was in fact correct and took a lot of heat for saying so. 40% is what they were getting when there were fab issues for a month or so.

The heat was well deserved and the article really just seemed like some Nvidia fud spin. It's really sad that the article was even posted on this web site.


What are you talking about?

The article I posted has to do with cypress yields being 60-80%.. Nowhere did it ever mention 4% anything..

Did I miss some areticle in the last couple of days stating cypress had 4% when we knew itr was close to 40% during the terrible issues times..???

I'm baffled and confused..
January 12, 2010 6:17:02 PM

TSMC is prime to lose business to GF if they cant get their ducks in a row.
a b U Graphics card
January 12, 2010 6:18:16 PM

Maybe it was third hand? meaning the writer got the news from someone who originally heard it from someone else, and they made the mistake, and the source gave the writer the wrong info, and wasnt corrected til it ran down to the original source?
Also, ATI has sold 2 million 5 series cards so far, and seeing Apes 2:1, thats the same ratio for the 5series vs the 58xx series, if you will, or, 1/3, near 700,000 58xx cards have been sold
The other article was from TG, and was a misprint/typo misunderstanding
a b U Graphics card
January 12, 2010 6:23:17 PM

JAYDEEJOHN said:
Maybe it was third hand? meaning the writer got the news from someone who originally heard it from someone else, and they made the mistake, and the source gave the writer the wrong info, and wasnt corrected til it ran down to the original source?
Also, ATI has sold 2 million 5 series cards so far, and seeing Apes 2:1, thats the same ratio for the 5series vs the 58xx series, if you will, or, 1/3, near 700,000 58xx cards have been sold
The other article was from TG, and was a misprint/typo misunderstanding



Ah I see..

Though this article from fud is claiming the cypress yields may be as high as 80%, and no lower than 60%. That could be a rather large gap in GPU price. 3:1, or 4:1.. not 2:1 based on 40%... though if we assume teh same issues that held ati to 40% are now hurting fermi to 20% we can expect 2:1 to be true come the spring. That is still a lot of lost profit given how close the cards will be in performance.. makes it even more likely we will see a 5890 in time for the 380 too..
a b U Graphics card
January 12, 2010 6:32:20 PM

Didn't Nvidia consider that wood screws are condustors of heat and can/will burn under the right conditions?

Meanwhile in cartoon-land, Intel is still suffering from malpractice with OEM's...
a c 130 U Graphics card
January 12, 2010 7:50:19 PM

JAYDEEJOHN said:
Maybe it was third hand? meaning the writer got the news from someone who originally heard it from someone else, and they made the mistake, and the source gave the writer the wrong info, and wasnt corrected til it ran down to the original source?
Also, ATI has sold 2 million 5 series cards so far, and seeing Apes 2:1, thats the same ratio for the 5series vs the 58xx series, if you will, or, 1/3, near 700,000 58xx cards have been sold
The other article was from TG, and was a misprint/typo misunderstanding


Yes i saw the thread yesterday i think it was and typo was the main reaction to it.

Now i need to pull you on the 2m 5 series cards sold there JD. What was said was that they had shipped, not sold 2m DX11 GPU, not 5 series cards.

Mactronix
a b U Graphics card
January 12, 2010 8:01:08 PM

True, but what was also said was it was 2:1, meaning the 58s had approx 700,000, and since we know between shipping and availibility, its thin, its pretty close, as theres still back order filling. or just shipped, not yet delivered purchases.
Anonymous
a b U Graphics card
January 12, 2010 8:07:29 PM

If Fermi is 50% bigger and 20% faster than a 5870 that means it hasn't improved on the g200 at all.

Still at least THG can do a 'Fastest single gpu card' roundup and get the desired result.

Much more of a concern is the 20% yields. That's pathetic, and you wont see any of these in desktops at that yield. They will all end up as Tesla's. The worst thing about that is, AMD have no reason to drop prices on any of their gpu's.
a b U Graphics card
January 12, 2010 8:13:47 PM

TheGreatGrapeApe said:
I think the 4% yields was a typo, even TSMC was saying 40% in December, I think 4% is not considered a yield, (about 2-5 chips per wafer). There is no way HD5870 was 4% because that would work out to about $500-1K per chip (not board, per GPU).

I think it's 20% vs 40%. which means it's still running at about 2:1 for chips cost between the two companies, just like during the GTX280/HD4K series.


I see. That was probably the case.
a c 130 U Graphics card
January 12, 2010 8:23:43 PM

JAYDEEJOHN said:
True, but what was also said was it was 2:1, meaning the 58s had approx 700,000, and since we know between shipping and availibility, its thin, its pretty close, as theres still back order filling. or just shipped, not yet delivered purchases.


No disrestect meant to either you or APE here JD but frankly i dont care about any 2:1. The article said 4% which was assumed to be a typo by many but it also said that while the yeilds would be low it wouldnt be worse than AMD/ATI. There are a lot of rumours going around and its hard to know what to beleive, although TGGA is usually about right.
Anyways 2:1 or not that dosent alter the fact that there is a hell of a lot of differance between actually selling 2m 5 series cards and just shipping 2m GPU. They could have 1m sitting in warehouses split between differant board partners for all knowing they have shipped then tells you.
Im not trying to start an argument here just want people to understand the huge differance between what you posted and what was actually said.
Guess the link wont work if its been taken down but for what its worth http://www.tomshardware.co.uk/nvidia-fermi-gpu,news-325...

Mactronix :) 
a b U Graphics card
January 12, 2010 8:28:34 PM

Problem I see is, having decent availiility not til April possibly, thats not far from a refresf from ATI, where if its close, a cooler, less power needing card doing almost as well 5890 could rain on it after only a few short sunny days/weeks, and pricing would be driven down, or itll come in low to begin with
a b U Graphics card
January 12, 2010 8:34:40 PM

mac, Im not referring to the 2:1 as Ape is, it only reminded me of what Id said.
My thoughts/posts come from wavey Davey, and reflect them as of earler this month.

Id bet that most are already made and shipped, and are sold, since availibility still isnt high, so we may be wondering about 10s of thousands here vs say 200,000.
Anonymous
a b U Graphics card
January 12, 2010 8:35:25 PM

Availability isn't going to be decent for this ever I feel. Look at how bad the 5800's were, demand for Fermi will be incredibly high like for any Nvidia card.

If Fermi is at 20% and Cypress 60-80%, that means it's not just TSMC who are messing it up.

If it's 20-40% slower than the 5970 but costing the same best case? Forget it, there is no point in even releasing it as Geforce.
a b U Graphics card
January 12, 2010 8:39:46 PM

Of 2 million, dividing by 3, 1 part are 58s, and those are mostly sold parts, hope this clears it up, and those numbers are accurate
a b U Graphics card
January 12, 2010 8:48:56 PM

Hey, i dont understand about half of what is being said here with the yields and such. any chance someone knows a link that explains it as im trying to learn a ton more about PC Hardware in general and stay up to date :) 

Also giving me sites to use for all the new info would be sweet as well.
a c 217 U Graphics card
January 12, 2010 8:50:49 PM

Essentially they are saying that it is rumered, by someone who works close with the plant making the GPU chips, that Fermi is having 20% yields. 20% is the number of chips made that pass specifications, or in other words, 20% of the chips made are able to be used in a working graphics card.
a b U Graphics card
January 12, 2010 8:53:08 PM

Yes, and that ATIs are at 40% or 2:1
My references are to sales only, and the 2:1 is 2 parts lower end 5xxx series vs 1 part 58xx series
a b U Graphics card
January 12, 2010 8:56:55 PM

Ah that explains a lot more and makes MUCH more sense. thank you. haha i was checking out fudzilla and saw the new EVGA best of a MOBO that fits into a pizza box and i think i wet myself :( 
a b U Graphics card
January 12, 2010 9:01:16 PM

Thats numbers ffom early decenber, not current.
This is from Dave Baumann from ATI

2 million shipped, 2:1 ratio as of recently
Anonymous
a b U Graphics card
January 12, 2010 9:12:02 PM

There have been various sources mentioning the 2 million shipped Cypress/Juniper parts.

You probably need to look at it logically and think at least 1.5 million of those are sold. Absolutely no way are ATI's partners stocking 1/2 million of them in warehouses - there are still plenty of places where you cannot get a hold of 5850's.

The UK for example. http://www.scan.co.uk/Shop/Computer-Hardware/All/GPU-AT...

One of the largest e-tailers still has pretty poor availability. They are selling and there is no sign of them stopping.
a c 130 U Graphics card
January 12, 2010 9:12:53 PM

JAYDEEJOHN said:
Thats numbers ffom early decenber, not current.
This is from Dave Baumann from ATI

2 million shipped, 2:1 ratio as of recently



Nvidia Fermi Production and Availability 08:50 - Monday 11 January 2010 by Tuan Nguyen
Source: Tom's Hardware US – Keywords: nvidia, fermi, gpu Category : Graphics Cards
Just a quick update regarding more information on Nvidia's upcoming Fermi GPU:

We received word from an Nvidia partner that Fermi will go into production starting in the 3rd week of February, and be available in "low quantities" starting mid-march.

According to our source, Fermi's yield will be low but not any worst than what AMD/ATI is experiencing. TSMC, the company producing Radeon HD 5800 series at 40nm, is "happy at 4-percent yield," said our source.

Nvidia's Fermi GPU will be the company's first DX11 GPU.

Source http://www.tomshardware.co.uk/nvidia-fermi-gpu,news-325...


Mactronix
Anonymous
a b U Graphics card
January 12, 2010 9:18:26 PM

Mac you and JDJ are talking about two different things, and two different articles I think.
a c 130 U Graphics card
January 12, 2010 9:21:48 PM

Quote:
Mac you and JDJ are talking about two different things, and two different articles I think.


That would be normal im talking about Fermi availability as is most everyone else as its what the thread is about. We got to ATI because JD posted something i thought missleading and this is wherewe ended up. :) 
a b U Graphics card
January 12, 2010 9:22:51 PM

Ive tried to sy this, SS got it, but hios info is old.
Im talking about sales, not yields.
a b U Graphics card
January 12, 2010 9:25:30 PM

I was just reminded by TGGA when he said about 2:1.

And yes, this is about a wrong article to begin with, with wrong numbers, and eventually the sales when they do get here, so I included whats been going on while everyones waiting
Anonymous
a b U Graphics card
January 12, 2010 9:25:50 PM

Yep JDJ is talking about ATI's dx11 sales, Mac is talking yields.

The 2:1 ratio JDJ is talking about is the ratio of 5700's to 5800's that ATI have shipped - so 1.33 million 5700's and 666 million 5800's.

The 4% number was a nonsense, Fud's numbers of 20% fermi, 60% (Cypress) - 80% (Juniper) are much more likely. I added the () stuff, as I assume that is what the difference in yields is about.
a c 130 U Graphics card
January 12, 2010 9:29:51 PM

JAYDEEJOHN said:
http://forum.beyond3d.com/showpost.php?p=1378969&postco...
someone needs to tell someone this isnt right, and since ive "heard" it was pulled from another site as well, maybe Toms really needs to pull it?
http://www.tomshardware.co.uk/news/Components,1/



Could have fooled me i have visited it 4 times tonight already :) 

As far as ATI and whats been made goes just do the math JD they said in this article http://www.tomshardware.co.uk/amd-gpu-dx11-radeon-direc...
That they released the 5 series 15 weeks ago They also say they have shipped tens of thousands of processors, not cards but processors, over the last few months.
Lets be charitable and give them the full 15 weeks at tens of thousands so thats 99,000 or else they would have said hundreds of thousands wouldnt they.
So does 99,000 X 15 = " million ? Answer no not even close.

Mactronix
a b U Graphics card
January 12, 2010 9:29:53 PM

Yields are double on ATI vs Fermi, or 2:1, and of those ATI numbers, 1 third are 58s, and are basically sold
a c 130 U Graphics card
January 12, 2010 9:35:02 PM

Hey thats cool, or not it edited what i quoted which at the time didnt include your edit hmm not sure if i like that or not as you cant see what was quoted in the first place. thus it could scew what was being said. Not in this case but in general i mean.

Mactronix
a b U Graphics card
January 12, 2010 9:38:21 PM

•The number of 5800 series video cards that have been shipped into the channel already is in the very "high tens of thousands."
http://www.hardocp.com/article/2009/11/10/o_5800_wheref...
Now, the making of them wasnt satisfactory back in November, at thousands per week then:
•AMD’s Add-in-Board (AIBs) partners are currently shipping thousands of 5800 series video cards per week on a worldwide basis.
Now, again, back in Nov, heres the projections:
•While currently there are thousands of video cards shipped on a weekly basis, we will likely see this weekly number reach into the "tens of thousands" come late November or early December. This hinges on whether or not AIBs keep air freighting video cards. The likelihood of air shipments to continue is very high, but AMD does not directly control this.

Now, since we all know that availibility has ramped way up since late Dec, as was said, those tens of thousands bacame hundreds of thousands

•Expect "well into the hundreds of thousands" of 5800 video cards to be shipped into the channel by year’s end.

So, what does well inoto the hundreds of thousands? 700,000?
a c 217 U Graphics card
January 12, 2010 9:38:26 PM

Given how radically different Fermi is from it's predessors, I wonder if it means the ~20% performance increase over Cypress (5870's if I understand correctly) is going to improve more than normal with driver revisions.
a c 130 U Graphics card
January 12, 2010 9:38:38 PM

JAYDEEJOHN said:
Yields are double on ATI vs Fermi, or 2:1, and of those ATI numbers, 1 third are 58s, and are basically sold


So is this part of teh article wrong as well then ? "According to our source, Fermi's yield will be low but not any worst than what AMD/ATI is experiencing."

If so then the writer needs dragging over hot coals, Reminds me though there was a post on here over the last few days concerning unsubstanciated claptrap with no source being posted as news and the guy got a flea in his ear for his troubles. seem she had a point after all.

Mactronix
a b U Graphics card
January 12, 2010 9:39:34 PM

Anyhow, thats whats Fermi is facing when it slowly dribbles out
a b U Graphics card
January 12, 2010 9:41:48 PM

mactronix said:
So is this part of teh article wrong as well then ? "According to our source, Fermi's yield will be low but not any worst than what AMD/ATI is experiencing."

If so then the writer needs dragging over hot coals, Reminds me though there was a post on here over the last few days concerning unsubstanciated claptrap with no source being posted as news and the guy got a flea in his ear for his troubles. seem she had a point after all.

Mactronix

I think its dated. TSMC announced 40% and higher awhile ago, and this 60% number isnt mentioned by FUaD for fermi, but is mentioned in a few places elsewheres for ATI

The only out he has is thru misinterpretation, meaning ATI experienced these low yields AT FIRST, and he can skate out on that
a c 217 U Graphics card
January 12, 2010 9:43:30 PM

mactronix said:
So is this part of teh article wrong as well then ? "According to our source, Fermi's yield will be low but not any worst than what AMD/ATI is experiencing."

If so then the writer needs dragging over hot coals, Reminds me though there was a post on here over the last few days concerning unsubstanciated claptrap with no source being posted as news and the guy got a flea in his ear for his troubles. seem she had a point after all.

Mactronix


If you reread the artical on the first post, they are suggesting that what has been rumored was incorrect, and the yields are not as good as are being said. The new source is supposed to be more accurate.

Both are not solid bits of info. It would seem Nvidia does not want the truth out, and wants to keep spinning things to keep interest high.
a b U Graphics card
January 12, 2010 9:44:29 PM

Apes right, its 2:1 yeilds, and maybe ranp as quickly as well, depending on how mature the process is at this point going forwards
Anonymous
a b U Graphics card
January 12, 2010 9:46:10 PM

Mac reading around the web on that THG yields article, it is getting panned for being really bad even by THG's standards.

http://www.semiaccurate.com/forums/showthread.php?t=134...

Whatever you think of charlie and SA, it's worth reading that.

I was also under the impression that the article was pulled btw.
a b U Graphics card
January 12, 2010 9:54:50 PM

It was pulled but it will be back with a correction.
a b U Graphics card
January 12, 2010 9:59:31 PM

Im loving this discussion thank you OP. I have learned a great deal about yields, wafer size and its correlation to pricing on GPU's. I find this very interesting so far.
a b U Graphics card
January 12, 2010 10:07:32 PM

What I wonder is, if the maturity of the process is good enough to offset the natural lower yields fermi will experience anyways, and itll see the same growth in yelds down the road?
a b U Graphics card
January 12, 2010 10:26:47 PM

bystander said:
Given how radically different Fermi is from it's predessors, I wonder if it means the ~20% performance increase over Cypress (5870's if I understand correctly) is going to improve more than normal with driver revisions.


Doubtful, you can say a lot of things about either company but nVidia's drivers tend to be better than ATI's on release and then they balance out in about a month or two.
a b U Graphics card
January 12, 2010 10:27:18 PM

JAYDEEJOHN said:
What I wonder is, if the maturity of the process is good enough to offset the natural lower yields fermi will experience anyways, and itll see the same growth in yelds down the road?


We don't know for sure what hampered the ATI yields.. It is possible that Nvidia will have to go through the same thing.
a b U Graphics card
January 12, 2010 10:31:48 PM

Thats true, tho a more mature process helps, tho only so far, if theres other problems, as you mentioned.

Id point out that the 20% perf is generally what we saw with the 4870 and the 280. Also, the driver "lead" was due to several things, one of which was nVidias having the early product, which isnt so this time around, as its fli flopped, as G80 was the first DX10 card by some time, as people waited on the 2900
!