iMacmatician
Regular
SemiAccurate: "Nvidia’s five new Keplers raise a red flag."
A1xLLcqAgt0qc2RyMz0y
Veteran
Which you don't.
You seem to think that nVidia held a gun to TSMC's head and forced a contract onto TSMC. And if you do then what you make AMD out to be when they had the same type of contract?
The "pay-for-good-die" is no longer available from TSMC for all parties not just nVidia and the increased costs of doing business with TSMC is being passed down to the buyers.
Just look at the 7970/680/690 prices.
A1xLLcqAgt0qc2RyMz0y
Veteran
Web site should be renamed MostyInAccurate.
That's not how it works. You pay for wafer starts. What you etch on the wafers is your business. How you design your chip has huge implications on yields -- so if you only pay for the working ones, you have little incentive to design chips that are less likely to be duds.
Blazkowicz
Legend
if nvidia raises red flags, that's fine 
No i dont, nv would force tsmc to sign a contract that was great for tsmc and bad for nv ????? doesnt make any sense
and why would you agree to pay for non working product ?
you in this case being nv, so your saying nv go round to tsmc's factory and do the etching ?
I beleived nv would create a design and tsmc would make it, both you and A1xLL say this is not the case
These kind of contracts (per good die or per wafer) are not as black and white as you make it out to be. In all cases, both parties will closely watch the actual yields and interact to root cause. This is never a throw-over-the-wall relationship. In addition, these contracts are often dynamic, with guaranteed minimum yields or minimum good dies etc. Also: the kind of yield failures impact price. Random failures all over the die are different than systemic failures in a localized area. In the latter case, parties need to determine who's responsible for the issue and price can be adjusted accordingly. Etc. etc.tunafish said:
Finally, even in a good die only situation, it is still very important to get high yields: for one the contract will often stipulate that price will go down per die if yields are good, but also, your wafer allocation will always be capped. So low yields will still hamper your ability to make money.
Somehow, Charlie makes it sound like the most evil thing to choose one of the offered contract options, where it was portraied as a successful business move, when AMD did execute on exactly the same model with GloFo last year.
http://www.zdnet.co.uk/news/process...balfoundries-wafer-supply-agreement-40092378/
http://www.zdnet.co.uk/news/process...balfoundries-wafer-supply-agreement-40092378/
Because it's being made according to your own plans.
The time period covered by the good die WSA was marked by a number of egg-on-face moments and a fair amount of public friction between the two partners over Llano's bad yields in particular.
Good-die did safeguard AMD from the downside of Llano's manufacturability problems by having GF eat additional costs if yields did not improve.
If Nvidia's situation is similar to AMD during that time period, it could be a bad thing for Nvidia.
AMD did not have a good time at 32nm during that time frame, something the arrangement with GF only partially obscured.
Not really, they seem fairly well priced where you'd expect for the performance. Checking newegg 580's are $380+. About right though you could quibble. 560's are priced ~50 less than 7850 for not too much less stock performance.
They are also discontinued products and I'm guessing the few remainders are not exactly flying off the shelves.
For current gen 680, it was actually underpriced at launch compared to 7970.
Now look at how they did in the market vs their competition.
Silent_Buddha
Legend
Sure but go back to GTX 580/Radeon 6970 launches up until 7970 came out.
GTX 580 was 500 USD while 6970 was 370 USD. A 130 USD price differential for a 10-15% perf increase.
I'd say that's a pretty obvious example. If you need more you can go back to the 5870 versus GTX 480. And that despite GTX 480 launching significantly later.
Or GTX 280/285 versus 4870.
It was pretty obvious Nvidia weren't going to adjust their enthusiast class (other than 280/285) due to price pressure from AMD. So AMD just adjusted their prices to be in line with Nvidia's.
Regards,
SB
I'd echo Silent_guy here. These are very complex, deep, strategic relationships. I'm sure there are calls between engineers at NV and TSMC on a weekly, if not daily basis to address current yields and future products. Working with a foundry simply isn't like buying web hosting, there is no 'fire and forget'.
When I look at the AMD/GF agreement, I largely see that as GF's way of dealing with the gate first screw-up on 32nm. TSMC simply isn't in a comparably bad situation.
Someone asked why Nvidia would go for 'wafer', rather than 'die' pricing. The answer is that TSMC may not have given them a choice. With a limited supply of wafers, TSMC can be more aggressive in pricing.
Also, if you think about it...a 'good die' pricing makes more sense when the process is immature and yields are fluctuating. It's an approach to encourage customers to adopt the new process and remove certain risks. OTOH, once the process has stabilized and is more mature, that's really not necessary.
Anyway, it's certainly true that Newegg is totally out of stock of the 600 series...so I suspect there is a kernel of truth to what Charlie is saying about Nvidia yields.
Honestly, I don't really expect a lot of 28nm from Nvidia till 3Q anyway...they tend to lag behind AMD by around 2 quarters in terms of process technology.
DK
When I look at the AMD/GF agreement, I largely see that as GF's way of dealing with the gate first screw-up on 32nm. TSMC simply isn't in a comparably bad situation.
Someone asked why Nvidia would go for 'wafer', rather than 'die' pricing. The answer is that TSMC may not have given them a choice. With a limited supply of wafers, TSMC can be more aggressive in pricing.
Also, if you think about it...a 'good die' pricing makes more sense when the process is immature and yields are fluctuating. It's an approach to encourage customers to adopt the new process and remove certain risks. OTOH, once the process has stabilized and is more mature, that's really not necessary.
Anyway, it's certainly true that Newegg is totally out of stock of the 600 series...so I suspect there is a kernel of truth to what Charlie is saying about Nvidia yields.
Honestly, I don't really expect a lot of 28nm from Nvidia till 3Q anyway...they tend to lag behind AMD by around 2 quarters in terms of process technology.
DK
seahawk said:
It's hard to tell on the internet whether people are being sarcastic, but just in case this post is serious: don't believe everything you read on Semiaccurate.
UniversalTruth
Veteran
Why do I think Charlie said 50 000 680s worldwide? 
Edit: I'm not sure if it was 50 000 but I think it's a five-digits number.
Edit 2: Why can’t Nvidia supply Kepler/GK104/GTX680?
In this article: "not more than 10000 worldwide to date".
But the comments following are rather interesting: "According to newegg sales history, it seems at least 10000 gtx680 are sold by newegg alone."
Edit: I'm not sure if it was 50 000 but I think it's a five-digits number.
Edit 2: Why can’t Nvidia supply Kepler/GK104/GTX680?
In this article: "not more than 10000 worldwide to date".
But the comments following are rather interesting: "According to newegg sales history, it seems at least 10000 gtx680 are sold by newegg alone."
Last edited by a moderator:
Charlie is always right.
