NVIDIA Kepler speculation thread

silent_guy said:
Looks like Qualcomm is having serious yield issues too. :LOL:
http://www.theverge.com/2012/6/13/3082481/qualcomm-snapdragon-s4-28nm-shortage-tsmc
Click to expand...

You do know that the word "Yield" is nowhere in the linked article!
Not enough "Supply" is mentioned though along with "shortages and fail to meet demand".

Is the "Smiley Face" meant that we should interpret your post as sarcastic (Like Universal wants us to treat his posts)?

If so I really don't see the need. Why not just post that Qualcomm believes TSMC's 28nm supply issues will continue until year end?


http://www.theinquirer.net/inquirer/news/2184090/qualcomm-expects-28nm-chip-supply-issues
 
Dave Baumann said:
I'm not sure why the :LOL: is there. Yes, that statement can point to a yield issue (or a postitive spin on a yield issues). TSMC has a captive market for 28nm right now and they have no need for excessive cap ex on 28nm production outlay beyond the planned demand from their customers; if their customers are saying there isn't sufficient supply then that mean they didn't plan the demand correctly or they need more wafers than they initially planned on due to yield, or a combination of both.
Click to expand...
Couldn't it also mean that their product is just really in demand past the number of wafer starts available in a reasonable time frame?
The phrase "plan the demand correctly" is open to some odd interpretations. Like Qualcomm may have screwed up by making its product too compelling.
 
Yes, absolutely, that is part of capacity planning. But all I'm saying is that one possability doesn't negate the other, and there is probably some factor of both there.
 
Dave Baumann said:
Yes, absolutely, that is part of capacity planning. But all I'm saying is that one possability doesn't negate the other, and there is probably some factor of both there.
Click to expand...
I agree it's probably a combination of both: yield ok but not yet at the full maturity stage and a couple of big companies fighting for the same bone.
 
So, after Nvidia now it's Qualcomm also who is not being able to satisfy demand due to limited capacity at TSMC. Put another way, both companies claim that they could sell more products than TSMC can deliver.

There's only one contract manufacturer for highly integrated ASICs that offers 28nm other than TSMC, right? So, AMD seems fine with their supply - could they have gone to Samsung with their manufacturing? ;)
 
Last edited by a moderator:
Somehow i ask me is if the problem is TSMC capacity or if brand dont ask them too much and over estimate the TSMC capacity ?

I think Samsung will really start to enter the fundry and manufacturing for nvidia, AMD etc ... they are just the funder with the higher capacity unit actually, and they have the technology.. They work with GF ( on 32nm production).. ( with the consortium made with the 4 productions unit of Samsung in USA and Dresden ( if im right )

There's not so much funder anyway: TSMC, GF... and Samsung ... ( Intel ofc too )
 
Last edited by a moderator:
I wrote this in another thread.

My card is acting weird, because before I installed the water block it actually boosted higher with stock settings or with any given off set setting. Now my card doesn't use the highest level of the voltage range 1.175v, unless I force it. In the past it pretty much defaulted to 1.175 at all times. I did change the Water block and of course changed the thermal paste and thermal gum or whatever that is called, so I don't know if that somehow affected things. I've heard it might be a driver issue, but doing a clean install didn't help me.

On my test today it did boost to higher level for a moment, but once I changed some settings I haven't been able to get that higher boost back. Even when I manually raise the volts to 1.175 I can't quite use the same clocks in 3dMark11, so lower temps don't do any good there.
Click to expand...

I've been discussing this issue with EVGA and just got word that they are aware of the issue and working on it. Basically the card throttles if temps go too low. It appears that this point is somewhere in the region of 43-44c as evidenced by my post. The temps were around there when it briefly boosted normally again. The problem is that in order for me to reach temps of 44c on the GPU I needed to simultaneously run Furmark and Prime 95 on a overclocked system and in a room with higher than normal ambient temperature... There is no way in hell I can reach that temp with regular gaming loads, unless I power off some radiator fans. Might have to try that actually.

Not really what I was looking for when decided to do water cooling, but it seems in the PC world, one just has to run into weird shit sometimes...
 
CarstenS said:
So, after Nvidia now it's Qualcomm also who is not being able to satisfy demand due to limited capacity at TSMC. Put another way, both companies claim that they could sell more products than TSMC can deliver.

There's only one contract manufacturer for highly integrated ASICs that offers 28nm other than TSMC, right? So, AMD seems fine with their supply - could they have gone to Samsung with their manufacturing? ;)
Click to expand...

Or they have caluclated the demand better and have signed up for a number of wafers at TSMC that fits the dmeand better.
 
seahawk said:
Or they have caluclated the demand better and have signed up for a number of wafers at TSMC that fits the dmeand better.
Click to expand...

Or put even more simply - there's lower demand for AMD's 28nm products compared to Qualcomm and nVidia. Crazy notion, I know.
 
Dave Baumann said:
At this point you are still coming through the planning phase. Its not BS, its a simple fact.
Click to expand...

What is a fact? The availability of 28nm parts in the retail channel is a simple case of demand trailing supply. That tells us nothing about how well AMD planned its wafer sourcing from TSMC. How do we know that you guys didn't overestimate demand?

Same goes the other way around. nVidia is obviously not meeting demand. Is that due to poor yields, poor planning or simply lack of capacity at TSMC? We have no numbers.

More specifically, how do you plan for high demand if your supplier is capacity constrained? Make the product less attractive? Increase prices (this has arguably already happened) ?
 
Dr Evil said:
I wrote this in another thread.



I've been discussing this issue with EVGA and just got word that they are aware of the issue and working on it. Basically the card throttles if temps go too low. It appears that this point is somewhere in the region of 43-44c as evidenced by my post. The temps were around there when it briefly boosted normally again. The problem is that in order for me to reach temps of 44c on the GPU I needed to simultaneously run Furmark and Prime 95 on a overclocked system and in a room with higher than normal ambient temperature... There is no way in hell I can reach that temp with regular gaming loads, unless I power off some radiator fans. Might have to try that actually.

Not really what I was looking for when decided to do water cooling, but it seems in the PC world, one just has to run into weird shit sometimes...
Click to expand...


Thats really damn strange, i know 670 had a limit set at 70°C before throttling, but i wasnt know this is a problem when lower temp is set.
 
lanek said:
Thats really damn strange, i know 670 had a limit set at 70°C before throttling, but i wasnt know this is a problem when lower temp is set.
Click to expand...

Yup a strange issue. I did some further testing and the throttling point seems to be pretty much exactly 43c on this card at least. Once the temp rises to 43c the card starts to use 1.175v and higher boost Mhz again. I hope this issue will be fixed soon.
 
trinibwoy said:
What is a fact? The availability of 28nm parts in the retail channel is a simple case of demand trailing supply. That tells us nothing about how well AMD planned its wafer sourcing from TSMC. How do we know that you guys didn't overestimate demand?

Same goes the other way around. nVidia is obviously not meeting demand. Is that due to poor yields, poor planning or simply lack of capacity at TSMC? We have no numbers.

More specifically, how do you plan for high demand if your supplier is capacity constrained? Make the product less attractive? Increase prices (this has arguably already happened) ?
Click to expand...

Honestly. Nvidia is selling so many parts that their suppliers can't keep up, and you're saying this is a bad thing??
 
keldor314 said:
Honestly. Nvidia is selling so many parts that their suppliers can't keep up, and you're saying this is a bad thing??
Click to expand...
That's not what he's saying. He's saying it can be taken either way -- good or bad. The good scenario being where Nvidia is selling so much product that TSMC can't keep up, or the bad scenario, where TSMC has such awful production of Kepler than Nvidia can't sell anything.
 
Yes... A random post somewhere on the web from someone who doesn't know what he's talking about always explains everything.
Have you wondered why for example there is a 7970 and 7950 on the list but on the other hand there is just "7800 Series"? This is exact same string as you see if you run for example dxdiag in the display tab under "name". Or the graphics card name that GPU-Z displays for example.

But of course an conspiracy theory involving NV paying millions of $ (that they get from selling their Kepler cards) to Valve just to hide the fact that they are really not selling well (so that they really don't have millions of $ to pay Valve in the first place) is much more likely. All we need now are some "sources" claiming that they seen their graphics card become a GeForce GTX 680 instead of GeForce 6100 for a second or two while steam survey decided to run and we have a semi accurate front page news!
 
You must log in or register to reply here.

Similar threads

A
RDNA3 Efficiency [Spinoff from RDNA4]
3 4 5
Replies
85
Views
7K
Bondrewd
Bondrewd
D
RTX Video HDR modded to work in games, RTX Game HDR?
Replies
7
Views
1K
DegustatoR
D
D
DX12, DX12U, DXR, API bias, and API evolution ..
2
Replies
31
Views
4K
trinibwoy
T
P
Speculation and Rumors: Nvidia Blackwell ...
18 19 20
Replies
392
Views
43K
Newguy
N
milk
Realtime AI content generation *spawn
2 3
Replies
49
Views
5K
Cyan
Cyan
Back
Top