News
 

Bookmark and Share

(15) 

Even though Nvidia Corp. commercially released its GeForce GTX 400-series roughly six months after the first public demonstration and had to disable parts of the chips to ensure higher supply, financial analysts from Needham and Company claim that Nvidia only has about ten thousand of Fermi-series chips on the market at the moment and that the yields of Fermi chips is between 20% and 30%.

“We are downgrading Nvidia to a “hold” and removing our 12-month price target (was $22) after a series of channel checks indicating that Fermi is not ramping well and there could be further product delays. Based on our findings, Nvidia has very limited supply of Fermi desktop/notebook parts and yields remain poor at around 20-30%,” analyst Rajvindra Gill wrote in a note to clients.

In a bid to improve yields of the code-named GF100 chip (NV60, GT300, G300), Nvidia disabled 32 out of 512 stream processors even on flagship GeForce GTX 480 graphics board, however, after the product was formally launched on the 12th of April, it is still rarely available and where it is in stock, it costs much more than its recommended price of $499.

Considering the fact that Nvidia GF100 chip – which powers both GeForce GTX 470 and GTX 480 graphics boards and will eventually power the Tesla C2000-series computing cards – consists of 3.1 billion of transistors and is made using TSMC’s 40nm process technology known for its insufficient yields, it is not a surprise that Nvidia cannot ship enough modern DirectX 11-supporting graphics processing units (GPUs) to compete successfully with its arch-rival ATI, graphics business unit of Advanced Micro Devices. Since Nvidia’s top-of-the-range GeForce GTX 480 has not managed to recapture performance leadership from ATI Radeon HD 5970 and is also not widely available, the halo effect of the advanced product launch will not be significant

The analyst from Needman believes that the potential delays of other chips from the Fermi family will inevitably mean additional design wins for ATI, which will lead to further market share loss by Nvidia. Recently ATI already announced that it had shipped over six milion DirectX 11 graphics processors since late September, meanwhile, according to Rajvindra Gill, Nvidia has supplied less than ten thousand of its GF100 chips.

“We believe Nvidia could lose market share starting in [the second half of 2010], face a more challenging pricing environment and/or experience potentially lower gross margins,” said the analyst.

Nvidia declined to comment on the information.

Tags: Nvidia, Geforce, Fermi, GF100, NV60, 40nm

Discussion

Comments currently: 15
Discussion started: 04/21/10 03:22:03 PM
Latest comment: 04/25/10 02:32:06 AM
Expand all threads | Collapse all threads

[1-10]

1. 
And Intel has no significant problems at 32nm.

Go figure.
0 0 [Posted by: Aristide1  | Date: 04/21/10 03:22:03 PM]
Reply

2. 
Everybody say it with me, "Charlie D, you were right!"
0 0 [Posted by: siuol11  | Date: 04/21/10 04:54:33 PM]
Reply
- collapse thread

 
Although not right on everything with this Fermi crisis, he was spot on about the extremely small availability of these cards. Also correct about the very low yields.

Low availability:
http://semiaccurate.com/2...gtx470480-launch-targets/
0 0 [Posted by: RtFusion  | Date: 04/21/10 08:14:26 PM]
Reply

3. 

And Intel has no significant problems at 32nm.

Go figure.



Yes, but Intels i7 980x has 1.17 billion
Fermi has 3 Billion, and on a bigger node.
0 0 [Posted by: ValiumMm  | Date: 04/21/10 05:27:03 PM]
Reply
- collapse thread

 
I think his point was that a bigger node should be less problematic... But Nvidia designed a chip that was way out of recommended spec for what started out as a troubled node, and it came back to bite them in the ass.
0 0 [Posted by: siuol11  | Date: 04/21/10 08:09:00 PM]
Reply

4. 
with ati re-punching nvidia in the face this year with new 6000 series on 40nm so no delays there, and moving down to 28nm on GF next year for 7000 series, I suddenly feel sorry for nvidia.

what can they do?
0 0 [Posted by: tabovilla  | Date: 04/22/10 01:16:28 AM]
Reply
- collapse thread

 
what can they do?


They can develop a GPU that doesn't suck. That would be a nice start.
0 0 [Posted by: FLA  | Date: 04/22/10 01:29:28 PM]
Reply

5. 
I don't think there's anyone that does not have doubts on TSMC's 40nm process, let alone upcoming process technologies. While it is true that Fermi has bigger die, CPUs run at higher frequency, and smaller nodes are harder to manufacture. Nvidia's problems with Fermi is both their fault + process tech.
0 0 [Posted by: DavidC1  | Date: 04/22/10 11:21:18 PM]
Reply

6. 
I think that the TSMC technology is pure crap. They are not even 1 quarter of liability and precision of Intel's fabs. 20% success doesn't admits ANY excuses. Not even those 3.5 billion transistors from the chip.
0 0 [Posted by: TAViX  | Date: 04/23/10 01:22:02 AM]
Reply
- collapse thread

 
You might just be forgetting something. Yields are not solely dependant on the quality of the manufacturing process; design is at least equaly important for the outcome. TSMC's 40nm tech is not flawless, but it has improved greatly over the course of the last year. This is evident when looking at the improved yields of Evergreen's derivatives.
Furthermore, stating that it is not even 20% of Intel's rate would imply that Intel is generating yields of 80-100% atm. Not sure where you got this info, but it would surely surprise me if this were true.
0 0 [Posted by: Mazlov  | Date: 04/23/10 05:45:37 AM]
Reply
 
Read again. I didn't say 20% of Intel's rate, I said exactly what the article was saying: 20% rate of success. So from 10 dies, only ~2 are good. But probably Intel has not 80% rate of success but around 50-60%, which is quite good.
0 0 [Posted by: TAViX  | Date: 04/25/10 02:32:06 AM]
Reply

7. 
For me nvidia fermi is better suited as a marketing tools rather than a product itself.

Nvidia need to have a fastest single gpu (though this is not the case, whereas 5970 beats them) in order for them to sell their crappy gpu... and yes that include gtx200-300 series.
0 0 [Posted by: ibmas400  | Date: 04/23/10 10:14:50 AM]
Reply

8. 
A highly speculative article and I bet they have at least one thousand units more in these past few days the article is up!

/sarcasm

Seriously tho, TSMC are in some deep guano at the moment and them announcing skipping on the next step process and going directly to smaller half-node isn't building anyone with much confidence considering how well 40nm is doing.
0 0 [Posted by: MyK  | Date: 04/24/10 05:18:51 AM]
Reply

9. 
So if you want to buy a fermi card. You're going to have to wait another 6 months.. go figure.
0 0 [Posted by: aethm  | Date: 04/24/10 06:53:38 AM]
Reply

10. 
So who's gonna compete with ATI/AMD Radeon, those evil neoliberals?
0 0 [Posted by: TeemuMilto  | Date: 04/24/10 09:54:05 AM]
Reply

[1-10]

Add your Comment




Related news

Latest News

Thursday, August 21, 2014

10:59 pm | Khronos Group to Follow DirectX 12 with Cross-Platform Low-Level API. Khronos Unveils Next-Generation OpenGL Initiative

10:33 pm | Avexir Readies 3.40GHz DDR4 Memory Modules. DDR4 Could Hit 3.40GHz This Year

12:10 pm | AMD to Lower Prices of A-Series APUs for Back-to-School Season. New Prices of AMD A-Series APUs Revealed

Wednesday, August 20, 2014

10:53 am | AMD to Cut Prices on FX-9000, Other FX Processors: New Prices Revealed. AMD to Make FX Chips More Affordable, Discontinue Low-End Models

10:32 am | LG to Introduce World’s First Curved 21:9 Ultra-Wide Display. LG Brings Curved Displays to Gamers, Professionals

9:59 am | AMD Readies FX-8370, FX-8370E Microprocessors. AMD Preps Two New “Mainstream” FX Chips