nVidia GT300 - GeForce GTX 380 yields are sub-30%

[Kakkoii]

BSN 6/24/2009
When we ran TSMC yield story, we left you owning an explanation. Time to clear that too? why TSMC has "f***** up" yields? Because the chip "some say it doesn't exist" has disastrous yields. We refuse to be drawn into the speculation does the GT300 exist or not, since we have enough data that would send nVidia Legal our way - but that was in the past, green guys learned better ;-)

According to our sources close to the [silicon] heart of the matter, the problem that nVidia has are yields in the 20 percentage range. You've guessed, that is waaay [insert several "a"] too low for launching volume production and we probably will see a new revision of silicon until the yields get high enough to earn a little bit of money. Having three faulty chips on one working one is much too much, since those faulty chips aren't exactly "GTX 360" or "slower Quadro FX" grade material. Some faulty parts might work under forced cooling, but the high leakage is an issue with the current graphics card layout. We won't go into the whole instability, does not work etc situation.

As we all know, the graphics chips are at the worst possible position, facing down [unless you put them in testbed/desktop case]. With high leakage parts, thermal shockwave is sent through the organic packaging to the PCB [Printed Circuit Board] and can cause extensive failure, like nVidia learned with their $200 million mistake called "bad bumps" or simply "bumpgate".

We have the exact number, but in order to protect the parties involved, we are going to refrain from posting the exact yield figure on first batch of chips. All we can say is - not yet ready for production.

Well, apart from the poor TSMC yields, it's promising to hear that Nvidia is already at this stage with GT300.

 

[Keysplayr]

Originally posted by: Kakkoii

BSN 6/24/2009
When we ran TSMC yield story, we left you owning an explanation. Time to clear that too? why TSMC has "f***** up" yields? Because the chip "some say it doesn't exist" has disastrous yields. We refuse to be drawn into the speculation does the GT300 exist or not, since we have enough data that would send nVidia Legal our way - but that was in the past, green guys learned better ;-)

According to our sources close to the [silicon] heart of the matter, the problem that nVidia has are yields in the 20 percentage range. You've guessed, that is waaay [insert several "a"] too low for launching volume production and we probably will see a new revision of silicon until the yields get high enough to earn a little bit of money. Having three faulty chips on one working one is much too much, since those faulty chips aren't exactly "GTX 360" or "slower Quadro FX" grade material. Some faulty parts might work under forced cooling, but the high leakage is an issue with the current graphics card layout. We won't go into the whole instability, does not work etc situation.

As we all know, the graphics chips are at the worst possible position, facing down [unless you put them in testbed/desktop case]. With high leakage parts, thermal shockwave is sent through the organic packaging to the PCB [Printed Circuit Board] and can cause extensive failure, like nVidia learned with their $200 million mistake called "bad bumps" or simply "bumpgate".

We have the exact number, but in order to protect the parties involved, we are going to refrain from posting the exact yield figure on first batch of chips. All we can say is - not yet ready for production.

Well, apart from the poor TSMC yields, it's promising to hear that Nvidia is already at this stage with GT300.

With all the powerful options currently available right now, it really isn't going to hurt much to have to wait for the new gen cards. As long as they are out in a reasonable amount of time for Windows 7 launch, I think we will all "be OK".

 

[Denithor]

Originally posted by: Keysplayr
With all the powerful options currently available right now, it really isn't going to hurt much to have to wait for the new gen cards. As long as they are out in a reasonable amount of time for Windows 7 launch, I think we will all "be OK".

Grumble. I'm waiting for GT300 so I can crank up my F@H ppd. Cannot come soon enough for me!

Besides which, doesn't low yields simply mean they will have a spread of "lesser" cards right from the start? Kinda like the 4830/4850/4870 from ATi - cores that didn't make the cut for the top end simply get sold for use in a lower model. Which is something severely lacking in todays nVidia lineup - GTX 260 isn't exactly the cheapest "low end" of a product series I've seen.

 

[VaultDweller]

With all the powerful options currently available right now, it really isn't going to hurt much to have to wait for the new gen cards.

Speak for yourself
I'm building a gaming rig not later than November. I'm really hoping that nVidia and AMD both release their next generation of cards by then. It looks like ATI will be first out the door, and that means their pricing will be through the roof while they wait for the competition to catch up.

 

[Paratus]

Don't worry I assume that TSMC will give NV their customary volume discount so that regardless of yields they will be hugely profitable for NV at anywhere from $200 to $650. Right?

 

[alyarb]

the 280 launched at $500 right? I don't see why GT300 has to be $650. They both started off with abysmal yields, except with GT200 nvidia wasn't gunning for a new fab process and *still* getting ~25% yields. Considering they are in this ballpark months ahead of launch, things can't be as bad as they were.

 

[TemjinGold]

Err... no. The 280 launched at $650. It dropped pretty quickly (2-3 weeks after launch) but I feel for those poor saps who got it at launch...

 

[OCGuy]

Originally posted by: alyarb
the 280 launched at $500 right? I don't see why GT300 has to be $650. They both started off with abysmal yields, except with GT200 nvidia wasn't gunning for a new fab process and *still* getting ~25% yields. Considering they are in this ballpark months ahead of launch, things can't be as bad as they were.

Reference 280 was $649 on launch day.


I think nV learned their lesson though. Unless 5XXX is a total performance failure, which I doubt, nV is not going to stress their partners with massive price drops again by pricing high. They lost XFX exclusivity due to the launch pricing debacle last year, and EVGA wasnt very happy either.

We are all getting spoiled right now with where things are priced. The prices of the next gen will be higher, but not ridiculous, in my opinion.

 

[yh125d]

I care more about the Top Gear nod (that was neat) than the bad yields. For some reason even though I've only ever owned nvidia, I'm becoming apathetic and pessimistic about them in general :

 

[OCGuy]

Originally posted by: yh125d
I care more about the Top Gear nod (that was neat) than the bad yields. For some reason even though I've only ever owned nvidia, I'm becoming apathetic and pessimistic about them in general :

This is not a nV exclusive issue with 40nm. Look into the 4770s.

 

[thilanliyan]

Originally posted by: yh125d
I care more about the Top Gear nod (that was neat) than the bad yields.

Where's the Top Gear nod?

 

[HOOfan 1]

Originally posted by: TemjinGold
Err... no. The 280 launched at $650. It dropped pretty quickly (2-3 weeks after launch) but I feel for those poor saps who got it at launch...

many of the board partners gave a refund to those who bought it at the higher price

 

[yh125d]

Originally posted by: thilan29

Originally posted by: yh125d
I care more about the Top Gear nod (that was neat) than the bad yields.

Where's the Top Gear nod?

Pic to the top-left of article. nVidia GT300: Some say... it doesn't exist. All we know it... it's [not] called the Stig.


<3 Clarkson

 

[deputc26]

Here's hoping they get it fixed :beer:

 

[nitromullet]

Lets assume (big assumption here) that GT300 is slated to launch Nov-Dec 09... If that is the case, how unusual would it be to see 20-30% yields for a new, huge chip on a new process 4 months before launch? Also, are they talking about yield for a chip intended for production with large quantities expected, or are they talking about low volume pre-production samples that may still undergo changes prior to production silicon?

 

[thilanliyan]

Originally posted by: yh125d
Pic to the top-left of article. nVidia GT300: Some say... it doesn't exist. All we know it... it's [not] called the Stig.


<3 Clarkson

Hehehe I see it now. Thanks. Oh and woohoo for Season 13 finally starting!!

 

[Idontcare]

Originally posted by: nitromullet
Lets assume (big assumption here) that GT300 is slated to launch Nov-Dec 09... If that is the case, how unusual would it be to see 20-30% yields for a new, huge chip on a new process 4 months before launch? Also, are they talking about yield for a chip intended for production with large quantities expected, or are they talking about low volume pre-production samples that may still undergo changes prior to production silicon?

Its not atypical or unusual by any means. If yields were >80% at this time then internally that would be viewed as meaning the design/layout schedule wasn't aggressive enough and they left time-to-market on the table.

Yields don't pay the bills, gross margins do. The question for NV is what price do they need to sell the chips at in order to pay their bills and then some.

Yields do directly feed into this, but if the chip delivers the performance then ASP will justify the diesize (and yields) by being commensurately larger as will gross margins.

I've seen single-digit yields be profitable to both foundry and design house, customers were willing to pay nearly $3k a chip so gross margins came in just fine.

 

[yh125d]

Originally posted by: thilan29

Originally posted by: yh125d
Pic to the top-left of article. nVidia GT300: Some say... it doesn't exist. All we know it... it's [not] called the Stig.


<3 Clarkson

Hehehe I see it now. Thanks. Oh and woohoo for Season 13 finally starting!!

Took em long enough. Clarkson should have won in the train though

 

[thilanliyan]

Originally posted by: yh125d
Took em long enough. Clarkson should have won in the train though

After all the flak he gives to trains!! No way.

Do you think it was Schumi driving in the lap of the FXX? Apparently there's only a couple of FXX cars without the white stripes...most have those stripes...and one of them is Schumi's (there's a video of him in it driving Zinedine Zidane around Magny Cours from a couple of years back) so maybe he requested to drive it (plus it blitzed the old record by like 7 secs which is pretty impressive)?

 

[nitromullet]

Originally posted by: Idontcare

Originally posted by: nitromullet
Lets assume (big assumption here) that GT300 is slated to launch Nov-Dec 09... If that is the case, how unusual would it be to see 20-30% yields for a new, huge chip on a new process 4 months before launch? Also, are they talking about yield for a chip intended for production with large quantities expected, or are they talking about low volume pre-production samples that may still undergo changes prior to production silicon?

Its not atypical or unusual by any means. If yields were >80% at this time then internally that would be viewed as meaning the design/layout schedule wasn't aggressive enough and they left time-to-market on the table.

Yields don't pay the bills, gross margins do. The question for NV is what price do they need to sell the chips at in order to pay their bills and then some.

Yields do directly feed into this, but if the chip delivers the performance then ASP will justify the diesize (and yields) by being commensurately larger as will gross margins.

I've seen single-digit yields be profitable to both foundry and design house, customers were willing to pay nearly $3k a chip so gross margins came in just fine.

Makes sense, although clearly NVIDIA isn't going to be able to sell GTX 380s for $3k

How realistic is the other scenario in my other post about samples? Do they do that, or is pretty much every chip made intended to be retail production quality, even though they know it will never happen? I guess what I'm asking is do companies actually make chips during testing, test them, and then re work them; or is all the design testing supposed to be done virtually prior to manufacturing?

 

[yh125d]

Originally posted by: thilan29

Originally posted by: yh125d
Took em long enough. Clarkson should have won in the train though

After all the flak he gives to trains!! No way.

Do you think it was Schumi driving in the lap of the FXX? Apparently there's only a couple of FXX cars without the white stripes...most have those stripes...and one of them is Schumi's (there's a video of him in it driving Zinedine Zidane around Magny Cours from a couple of years back) so maybe he requested to drive it (plus it blitzed the old record by like 7 secs which is pretty impressive)?

I'm not sure, I think it was all a stunt really. I don't think he's The Stig. I think they just had him portray the stig so people would stop prodding and speculating. They did the FXX part to give it a bit more credibility. I think Schum in the stig suit looked a bit smaller than when we've seen Stig before, lighter


I didn't think I'd be rooting for the train, but I was the whole episode. Too bad the vincent didnt fare better

 

[thilanliyan]

Originally posted by: yh125d
I'm not sure, I think it was all a stunt really. I don't think he's The Stig. I think they just had him portray the stig so people would stop prodding and speculating. They did the FXX part to give it a bit more credibility. I think Schum in the stig suit looked a bit smaller than when we've seen Stig before, lighter


I didn't think I'd be rooting for the train, but I was the whole episode. Too bad the vincent didnt fare better

Oh yeah no way Schumi is the real stig (lol he has way more important things to do)...the real one was outed a couple of months ago...I didn't want to know so I didn't bother checking out who he was but apparently he has some racing experience (not in F1 though).

But maybe in the FXX lap it was Schumi driving since I think that's his own car.

I was definitely cheering for the car. Those old Jags are really nice.

 

[Scali]

What I wonder though... all these rumours about poor 40 nm yields revolve around TSMC.
nVidia has also partnered with UMC, but you don't hear anything from them.
So even though TSMC may still be struggling with 40 nm (although officially TSMC, nVidia and AMD declared that the problems were solved), that doesn't necessarily mean that all 40 nm products will be plagued by yield problems. UMC may be their saving grace.

 

[Kakkoii]

Originally posted by: nitromullet

Originally posted by: Idontcare

Originally posted by: nitromullet
Lets assume (big assumption here) that GT300 is slated to launch Nov-Dec 09... If that is the case, how unusual would it be to see 20-30% yields for a new, huge chip on a new process 4 months before launch? Also, are they talking about yield for a chip intended for production with large quantities expected, or are they talking about low volume pre-production samples that may still undergo changes prior to production silicon?

Its not atypical or unusual by any means. If yields were >80% at this time then internally that would be viewed as meaning the design/layout schedule wasn't aggressive enough and they left time-to-market on the table.

Yields don't pay the bills, gross margins do. The question for NV is what price do they need to sell the chips at in order to pay their bills and then some.

Yields do directly feed into this, but if the chip delivers the performance then ASP will justify the diesize (and yields) by being commensurately larger as will gross margins.

I've seen single-digit yields be profitable to both foundry and design house, customers were willing to pay nearly $3k a chip so gross margins came in just fine.

Makes sense, although clearly NVIDIA isn't going to be able to sell GTX 380s for $3k

How realistic is the other scenario in my other post about samples? Do they do that, or is pretty much every chip made intended to be retail production quality, even though they know it will never happen? I guess what I'm asking is do companies actually make chips during testing, test them, and then re work them; or is all the design testing supposed to be done virtually prior to manufacturing?

This is likely one of their test revisions. It will most likely go through another revision to be retail. BSN also suggested this in their article.

 

[Idontcare]

Originally posted by: Kakkoii

Originally posted by: nitromullet
Makes sense, although clearly NVIDIA isn't going to be able to sell GTX 380s for $3k

How realistic is the other scenario in my other post about samples? Do they do that, or is pretty much every chip made intended to be retail production quality, even though they know it will never happen? I guess what I'm asking is do companies actually make chips during testing, test them, and then re work them; or is all the design testing supposed to be done virtually prior to manufacturing?

This is likely one of their test revisions. It will most likely go through another revision to be retail. BSN also suggested this in their article.

Well they don't intentionally make "test revisions". When they tapeout the chip it is supposed to be production ready, if it isn't then they don't tape it out (generate the masksets needed to produce the wafers).

Now it is common that bugs are uncovered in the design only once they have silicon in hand. So while the intentions are for first silicon to be production ready, invariably there will be some degree of fatal flaws uncovered which require correction and validation followed by yet another tapeout (called a respin).

So as nitromullet says, all design testing (verification and validation) is supposed to be done prior to manufacturing. It is simply too costly and too time intensive to rely on silicon samples as your feedback loop.

Now when they do a tapeout on a new design they will "hot lot" a lot of wafers (not a lot as in more than a whole bunch, but in fab-speak a "lot" of wafers is a group of wafers moved around the fab in things called foups, each "lot" may be 12 to 24 wafers, like a loaf of bread) to get samples out very quickly so they can verify whether the existing design is truly ready for production, and if it is then they open the floodgates on the remaining lots waiting back at lotstart.

Should the results from those first few wafers and chips come back as indicating some fatal flaws exist in the design then they simply won't make any more of them with those masks until the errors are located and corrected.

Or if the yields are so low that the gross margins aren't expected to be viable then they will wait for the yields to be fixed before making a bunch of unsellable chips.