NV to leave IBM?

[radar1200gs]

Bjorn, Yeah, apparently it has taped out. The real question is has it started ramping into production? If not, then it can be taped out elsewhere (after redesign for the design rules of the new fab) and only costs nVidia the price of the IBM tapeout.

 

[hstewarth]

Misinformation???

This sounds like misinformation to me. With information be published that NV40 is made by IBM, I think this story was dream up by inquirer to propose productions problem with NVidia's new chip.. Just a way attract information.

In all, until its official - it is meanless.

 

[Arun]

Dificult to say, but didn't Uttar say that the NV40 was the smotthest chip delivered to Nv? I'll believe it when i see something official, till then just rumors.

I never said that. I said the NV36 was the smoothest chip in NVIDIA's history. Not NV40. Also, that was several months ago - if yields didn't improve much at IBM, but improved much at TSMC, the situation might very well have reversed.
Also, radar1200gs, the NV38 is fabbed at TSMC. Not IBM.
Furthermore, TSMC's extremely spectatular success on Low-K - a technology NVIDIA was looking foward for a long time now - might have interested them also.

Uttar

 

[Megadrive1988]

This sounds like misinformation to me. With information be published that NV40 is made by IBM, I think this story was dream up by inquirer to propose productions problem with NVidia's new chip.. Just a way attract information.

yup, this i can believe

Inq *needs* there to be something *JUICY* in everything

 

[Qaz]

NV to leave IBM...

Few points from IBM Q1/04 conference call that may help:

1. A number of partners seem to still believe in IBM's semi business:

"In the first quarter two industry leaders, Sony and Samsung, attested to the value of IBM's technologies with strategic commitments. Sony Group is investing $325 million to facilitate the production of the next generation 65 nm chips in IBM's 300 mm semiconductor manufacturing facility. Samsung Electronics joined IBM's Chip Development Alliance with Infineon and Chartered Semiconductor, broadening the base of firms utilizing similar processes."

2. US$100M of stated US$150M loss was due to IP not yield problems:

"On a pro forma basis our former Technology Group would have lost approximately $150 million in the first quarter, of which roughly $100 million was due to IP. As I mentioned earlier, we announced a deal with
AMCC which is expected to deliver approximately $200 million of IP income in the second quarter. This will help the group's profit in the second quarter."

3. No sign of 'fleeing customers"

"...we expect the former Technology to be profitable in 2004 based on improving yields, flat IP income and increased demand."

4. Finally regarding the yield issues..

"First of all, as I mentioned, the 200 mm business in Burlington, they had a pretty good quarter with yields better than we thought. So that was good. Our 300 mm factory did see improved yields, just not fast enough. We are very focused on this. We have what are called lots that are moving through the factory where the yields are where we would like them to be. The problem is we can't get all of the lots there. But once you
start to have some successes you're able to figure out those fixes
and you can get the yields that we do expect."

Hope this info helps...

 

[Hellbinder]

What I don't understand: if TSMC has noticable advantages over IBM, why didn't nVIDIA let TSMC build the chip first? I heard IBM having problem for a long time, nVIDIA must knew it.

If you recall, nVidia blamed TSMC for all of their yield woes on NV30.

Exactly.. Which was mostly a lot of Rubbish!

Nvidias Nv30 woes were caused by their own engineering failures. But of Course Nvidia did again what they always do. Smear everyone else, Lie, Spread FUD and Deny everything..

 

[ZenThought]

What I don't understand: if TSMC has noticable advantages over IBM, why didn't nVIDIA let TSMC build the chip first? I heard IBM having problem for a long time, nVIDIA must knew it.

If you recall, nVidia blamed TSMC for all of their yield woes on NV30.

Exactly.. Which was mostly a lot of Rubbish!

Nvidias Nv30 woes were caused by their own engineering failures. But of Course Nvidia did again what they always do. Smear everyone else, Lie, Spread FUD and Deny everything..

You can't compare ATI's smaller chip with NVDA's much larger
NV30. NVDA was trying to push limits of 130 nm design at much
earlier stage when TSMC probably having lots of difficulity.

You are spreading FUD again...

 

[Saem]

Counting metal layers isn't straight forward, in that it's ambigious. There are two methods of counting.

 

[JoshMST]

From my understanding, bulk 130 nm process at IBM is solid. When you get to 90 nm and 130 nm SOI, things look like they are not so peachy. Transistor performance for bulk IBM is slightly better than bulk 130 nm TSMC, but performance of TSMC low-k is better than IBM bulk 130 nm.

Also, near the end of the NV30 development process, TSMC changed their 130 nm libraries dramatically to deal with problems in the 130 nm process. This is rather late to do once the design is nearly finished. Then there are the rumors that TSMC told NV that low-k will be ready by summer 2002. NV did in fact mess up because I believe initial clockspeeds on NV30 were set to be around 400 MHz, and trying to get it over that took the infamous cooling.

Geeze Hellbinder, such vitriole. NV is no better or worse than any other tech company I know. They just have a very high profile. The smear that AMD and Intel do to each other makes what ATI and NV do to each other look petty.

 

[CyFactor]

You can't compare ATI's smaller chip with NVDA's much larger
NV30. NVDA was trying to push limits of 130 nm design at much
earlier stage when TSMC probably having lots of difficulity.

You are spreading FUD again...

Pray tell, why, then, has ATI successfully produced several .13 micron low-K parts wheras nVidia has yet to produce one? R420 is also low-K, incidentally (if you believe the rumors). And that's hardly "small".

You are spreading FUD again.

 

[Mindriot]

You can't compare ATI's smaller chip with NVDA's much larger
NV30. NVDA was trying to push limits of 130 nm design at much
earlier stage when TSMC probably having lots of difficulity.

You are spreading FUD again...

Pray tell, why, then, has ATI successfully produced several .13 micron low-K parts wheras nVidia has yet to produce one? R420 is also low-K, incidentally (if you believe the rumors). And that's hardly "small".

You are spreading FUD again.

Correct me if I'm wrong but the 9600XT (RV360) is a 130nm low-k part isn't it?

 

[ChrisRay]

I do wonder sometimes if people Realise, its harder to get a working chip from a 125m transistor part than a 70m transistor part (Could be off on the Rv350 but its close still) With 0.13 tech

 

[CyFactor]

Correct me if I'm wrong but the 9600XT (RV360) is a 130nm low-k part isn't it?

Correct. So is M11, AKA Mobility 9700. So is the RV380 (from the rumor mill). And so is the R420 as well as all of its variants (also from the rumor mill).

That's my point. If all the problems nVidia experienced were because of 130nm low-k at TSMC, why have they yet to produce a part on this process even after ATI has successfully demonstrated that it works and isn't "dangerous"?

 

[ZenThought]

Correct me if I'm wrong but the 9600XT (RV360) is a 130nm low-k part isn't it?

Correct. So is M11, AKA Mobility 9700. So is the RV380 (from the rumor mill). And so is the R420 as well as all of its variants (also from the rumor mill).

That's my point. If all the problems nVidia experienced were because of 130nm low-k at TSMC, why have they yet to produce a part on this process even after ATI has successfully demonstrated that it works and isn't "dangerous"?

Read previous posts. RV380 is 60-70m transistors. all NVDA's
chip was designed probably at 2001-2002. ATI's chips was designed
around 2002-2003. After problem with TSMC, NVDA decided to
fab at IBM thus skipping TMSC's low-k.

 

[CyFactor]

I do wonder sometimes if people Realise, its harder to get a working chip from a 125m transistor part than a 70m transistor part (Could be off on the Rv350 but its close still) With 0.13 tech

Considering R420 is on the order of 180+m transistors and working on low-k (allegedly)... Now getting it working a year ago may have been quite a feat, but my point is nVidia still has not produced a part successfully using this process. Who knows what NV45 will do, though.

And it's not just a matter of getting it working; it's a matter of yielding enough parts at target speeds for production quantities.

 

[martrox]

Also remember, in 2001, that TSMC had warned both nVidia and ATI that both .13 and low-k wasn't ready, and recommended not using it. ATI took them at their word, hence the .15 R3xx series.

 

[ChrisRay]

I do wonder sometimes if people Realise, its harder to get a working chip from a 125m transistor part than a 70m transistor part (Could be off on the Rv350 but its close still) With 0.13 tech

Considering R420 is on the order of 180+m transistors and working on low-k (allegedly)... Now getting it working a year ago may have been quite a feat, but my point is nVidia still has not produced a part successfully using this process. Who knows what NV45 will do, though.

And it's not just a matter of getting it working; it's a matter of yielding enough parts at target speeds for production quantities.

Well they are using IBM now for their process, Why are we discussing Low K? Nv30 originally wanted the feature, But they canned support for it. And Yields were still Low, Correct?

I am sure 13 tech has improved alot. I mean NVidia is now making NV35s by the bundle with more Transistors than the NV30, That couldnt be said for a year ago, looks like ATI playing 0.13 tech on low end chips was a bright Move,

But it doesnt mean that 0.13 tech was ready for such a big transistor chip at the time.

 

[CyFactor]

Well they are using IBM now for their process, Why are we discussing Low K? Nv30 originally wanted the feature, But they canned support for it. And Yields were still Low, Correct?

...

I started the low-k discussion as an example of how ATI had pushed the limits of process technology successfully to rebuke ZenThought's post:

You can't compare ATI's smaller chip with NVDA's much larger
NV30. NVDA was trying to push limits of 130 nm design at much
earlier stage when TSMC probably having lots of difficulity.

You are spreading FUD again...

In a sense, throwing FUD to counter his FUD since he showed dissent about spreading FUD.

 

[{Sniping}Waste]

Ill say it again. The IC layout can make good yeilds or bad yeilds. The IC layout is one of the bigest keys in yeild. The big compaies that jumped on to auto routing is paying for it now. Intel is a good exsample. Micron does not alow any auto routing and the yeilds are in the 90% range because of it. It also alow small changes with out having to re-route to wole IC. For the most part Micron IC for the most part work on the first steping for production. The problem for the most part is Nvidia and part IBM. If you have a poor IC layout than the best FAB will still give you yeild problem.

(joke) maybe somone in IBM is adding salt in the HCL tanks.

 

[Ailuros]

Megadrive:
It was painless compared to TSMC. Look at nVidia's former cores from TSMC. Most of them required at least 3 respins before being released to production (GF3 was A4 at release). NV40 is going to be released on A1 or A2 silicon (NV38 & NV36 were just as good). That saves a lot of money and time.

Did you also count A0x silicon revisions in order to make such comparisons?