Nvidia Kepler Yields Lower Than Expected –CEO. Fermi 2.0?

[Ajay]

The 7950 is the same chip, just with functional units disabled/not working so I do assume they are die harvesting.

Not necessarily. ATI made a shift in strategy back in 2005 and decided it wasn't in the best interests of the company, or customers to use die harvesting or repairability: http://www.anandtech.com/show/2679/3

Whether or not they are sticking to that line of thinking for the 79XX chips - I don't know.

 

[wahdangun]

That was a mock up --- there was fermi hardware running for the GTC Demo they showcased that day.

can you show me the hardware that was used in GTC Demo ??

hell even intel used VLC player to show of their graphic capability. lol

 

[SirPauly]

I'm listening to the conference call now (thanks for the link, not) and JHH said several times that they are working with TSMC and that yields will improve. Which is stating the obvious to say the least of course they will. AMD is somehow able to ship 28nm in good quantities, so did AMD get all the good wafers somehow? JHH said that hardly anyone is shipping 28nm products, even though their direct competitor has been shipping them since last year. Quite disingenuous on his part.

Nvidia's problem is NOT that they need more wafers, their problem is they can't produce Kepler on the wafer that they have. Otherwise we would see Kepler now, just in short supply.

Hopefully this will all go away soon I'm tired of waiting. D:

Maybe AMD and nVidia define yields and good quantities differently for their 28nm products.

Maybe nVidia desires to bring more performance/value to the consumer.

 

[Ajay]

With the prices of the GTX 580 staying fairly high and the prospects that its successor (performance wise) will need to be priced high because of supply constraints, the chances of me upgrading this summer to reach that level of performance are approaching zero

 

[AnandThenMan]

AMD is the only company which has 28nm products from TSMC on the market. So if you based the state of the process on this than it seems that every other customer has problems...

From TSMC:

Marced emphasized that momentum by saying that TSMC now has 36 ICs in production and 132 tape-outs in the pipeline. "The momentum on 28-nm is three times what it was on 40/45-nm," she said.

source

Is one of those Kepler?

 

[Arzachel]

I think they will launch BigK and maybe GK104 for desktop and put every GK106/7 Chips into the OEM market.

The same wafers are used for all of those. If they are missing targets for OEM's, you won't see many GK104's being made.

Design issue. The power consumption was so high that i think they had no real chance to bring the full GF100 to the market. Even with GF110 its a problem.

Reread the amazingly informative posts IDC made earlier in this thread. Missing clockspeeds and voltages might come down to Nvidia having made too optimistic projections instead of parametric yield, but the locked shaders definetely can't. You don't lock shaders unless they absolutely don't work or you want to bin the chip. Else downclocking would've worked far better.

I know using exact quotes and data from the companies surly doesn't make much sense compared to rumor mongering, unnamed sources, conjecture, speculation and Internet Lore.

I risk sounding like a bit of a dick and undermining my point, but we're on the Internets, hyperbole is what we are, so here goes:

You might prefer listening to "Dear Leader" saying everythings just fine, but I look at a product that is late, crippled and has next to no supply and can't help but think that something is amiss. (Thats about the launch of the GTX480, now I just look and see nothing.)

 

[notty22]

I'm listening to the conference call now (thanks for the link, not) and JHH said several times that they are working with TSMC and that yields will improve. Which is stating the obvious to say the least of course they will. AMD is somehow able to ship 28nm in good quantities, so did AMD get all the good wafers somehow? JHH said that hardly anyone is shipping 28nm products, even though their direct competitor has been shipping them since last year. Quite disingenuous on his part.

Nvidia's problem is NOT that they need more wafers, their problem is they can't produce Kepler on the wafer that they have. Otherwise we would see Kepler now, just in short supply.

Hopefully this will all go away soon I'm tired of waiting. D:

You are guessing and offering a opinion.
This would seem to show AMD does not have enough 28nm wafers either.
AMD's 200-450 dollar 28nm product is in the same place as Kepler, not for sale yet.
Radeon HD 7700-series availability is spotty

Based on our findings, Newegg seems to have gotten the bulk of initial shipments. Amazon has not one card in stock. An unlabeled MSI Radeon HD 7770 is available through the Amazon Marketplace, but at $200.64, it's prohibitively expensive. Meanwhile, TigerDirect and NCIX each have all of three cards ready to go.
The Radeon HD 7750's availability situation is even more dire. Listings are hard to come by, and we only found a single model in stock. At least it wasn't marked up above AMD's suggested e-tail price.
As we said in our review, the Radeon HD 7770 and 7750 are both going to need price cuts before they become truly appealing. Right now, the Radeon HD 7770 is especially hard to recommend when the quicker Radeon HD 6850 can be nabbed for as little as $139.99. 7700-series prices will no doubt go down over time—and more cards will no doubt hit stores—but there's a lot of room for improvement right now

 

[SirPauly]

can you show me the hardware that was used in GTC Demo ??

hell even intel used VLC player to show of their graphic capability. lol

http://www.youtube.com/watch?v=fYuH2Kl_b98&feature=related

5 minute mark.

 

[notty22]

can you show me the hardware that was used in GTC Demo ??

hell even intel used VLC player to show of their graphic capability. lol

What a helpful post.

 

[SirPauly]

The same wafers are used for all of those. If they are missing targets for OEM's, you won't see many GK104's being made.



Reread the amazingly informative posts IDC made earlier in this thread. Missing clockspeeds and voltages might come down to Nvidia having made too optimistic projections instead of parametric yield, but the locked shaders definetely can't. You don't lock shaders unless they absolutely don't work or you want to bin the chip. Else downclocking would've worked far better.



I risk sounding like a bit of a dick and undermining my point, but we're on the Internets, hyperbole is what we are, so here goes:

You might prefer listening to "Dear Leader" saying everythings just fine, but I look at a product that is late, crippled and has next to no supply and can't help but think that something is amiss. (Thats about the launch of the GTX480, now I just look and see nothing.)

He didn't say everything was fine, he offered he would like to see the yields increase and would like more wafers. The guidance for margins has been reduced to around 49.2. What are you reading?

 

[Keysplayr]

@Keysplayer

::snip:: personally, i'd have liked a lower price ::snip::

All I needed to know. And thank you.

 

[sontin]

From TSMC:

Is one of those Kepler?

Sure.

So there are 36 "IC" in production but only 2 on the market?

 

[AnandThenMan]

http://www.youtube.com/watch?v=fYuH2Kl_b98&feature=related

5 minute mark.

Where do you actually see the hardware?

 

[Arzachel]

He didn't say everything was fine, he offered he would like to see the yields increase and would like more wafers. The guidance for margins has been reduced to around 49.2. What are you reading?

If their yields are down they need proportionately more wafers to ship the same volume. Don't mistake what I say for doomsaying, this is going to be only mildly inconvinient at worst for Nvidia, but you won't see them competing on price, thus the chance of prices going down for discrete desktop GPU's overall are close to nil in the coming months. Which sucks for me as a consumer.

"fabulous node" could mean a lot of things. It could mean 28nm is great in theory, will be great, great for what it promises. It doesn't automatically mean Nvidia is having great success with it currently.

If I had a hint of artistic talent, I'd totally draw a "fabulous node"! ...fabulous

 

[AnandThenMan]

So there are 36 "IC" in production but only 2 on the market?

I don't know. So AMD beat everyone else to 28nm?

 

[sontin]

I don't know. So AMD beat everyone else to 28nm?

So nVidia has no problem with 28nm?

I mean we know that Qualcomm is using 28nm, too. And there is not even a date for products...

 

[wahdangun]

http://www.youtube.com/watch?v=fYuH2Kl_b98&feature=related

5 minute mark.

lol, just check this http://www.youtube.com/watch?feature=player_detailpage&v=fYuH2Kl_b98#t=235s

he even blatantly lied, he said it was REAL FERMI. but in the end it was nothing more than

he even said it was fresh from fab. lol the level of fail in that video is so strong

EDIT: don't tell me you believe it was real fermi card in that video ??

 

[lifeblood]

If I was building/selling a product like GPUs and was supply constrained (planned or not), I would sell what I had at a high price. Because my price is so high I would have a demand equal to my supply. I would also have an incoming cash flow. By releasing the GPU I am getting it in the hands of developers and users who are giving me feedback and allowing my driver writers to fix problems, and application writers the chance to write apps that use my product.

I am certainly badly over simplifying things with this example, but conceptual it sounds like what AMD is doing. They are first to market and receiving all the benefits (tangible or not) from that. Maybe tomorrow NV will release their products that will deliver a knockout punch to the 7000 series, but today AMD is in the ring alone.

 

[Arzachel]

So nVidia has no problem with 28nm?

I mean we know that Qualcomm is using 28nm, too. And there is not even a date for products...

Well you could say that AMD's node change is going really well and Nvidia's is going "normal"... but they are direct competitors. If one of them is doing better, then the other is doing worse and vice versa.

 

[AnandThenMan]

Maybe AMD and nVidia define yields and good quantities differently for their 28nm products.

Perhaps.

Maybe nVidia desires to bring more performance/value to the consumer.

Hopefully it's worth the wait.

I finished listening to the conference call, Jensen was asked at least 3 times about 28nm, and was asked specifically about 28nm graphics products. The questions were obviously to elicit exactly when Kepler would ship, but JHH didn't blink and stuck to his, "we need more wafers" line. So there is nothing to take away from that call when Kepler will be available. Later today we may know more.

 

[sontin]

Well you could say that AMD's node change is going really well and Nvidia's is going "normal"... but they are direct competitors. If one of them is doing better, then the other is doing worse and vice versa.

How is AMD hurting nVidia when their problem is 28nm and not the competition?

Q1 is the first real quarter of 28nm shipments. nVidia is building inventory and will ship them for revenue. So it looks that the real enemy is the process.

 

[SirPauly]

lol, just check this http://www.youtube.com/watch?feature=player_detailpage&v=fYuH2Kl_b98#t=235s

he even blatantly lied, he said it was REAL FERMI. but in the end it was nothing more than

he even said it was fresh from fab. lol the level of fail in that video is so strong

EDIT: don't tell me you believe it was real fermi card in that video ??

Believe what ever you want. Makes no difference to me.

 

[formulav8]

AMD has always outclassed nvidia since they bought ATI on shrinks. AMD definitely have the right guys for new nodes. Maybe AMD's fab experience helped the ATI guys or something. I'm sure AMD helping TM also ends up helping nvidia.:thumbsup:

 

[TC777]

I don't know. So AMD beat everyone else to 28nm?

ATI/AMD seems to always beat Nvidia in technology, whether a card is faster or not, they always adopt new standards faster and use newer technology sooner.

 

[LOL_Wut_Axel]

I don't think GK104 will be that late, but if I'm honest the people mentioning GK110 launching in April are extremely delusional. It shouldn't prove hard for NVIDIA to launch GK104 graphics cards in April even if they're in limited quantities, but as it is not there's no working samples of GPUs based on GK110. A Q3 or Q4 2012 estimate for it seems just about right.