GK104 Kepler gpu's Defective?

No they should bring out real Kelper instead of hopped up GF104. Something worthy of upgrade. No offense GTX 680 owners but you only got 25-30% boost compared to over 60% from 280 to 480 and the list goes on.

At least they got it at the same price, unlike AMD owners. I'd rather have 30% more performance for the same price than 50% more performance for 60% more money.
 
At least they got it at the same price, unlike AMD owners. I'd rather have 30% more performance for the same price than 50% more performance for 60% more money.

True dat. And AMD did not give 50% either for their pathetic price increases. More like same 30% boost down the line from 68xx to 78xx and 69xx to 79xx. This is one area nV is pretty cool. Except that GTX 670 seems higher than 570 ever was...
 
DASHIT - I will say Charlie said nV was loosing tones of money and poor yields on fermi too and in reality made tones of money during Fermi era...sound familiar...
 
No they should bring out real Kelper instead of hopped up GF104. Something worthy of upgrade. No offense GTX 680 owners but you only got 25-30% boost compared to over 60% from 280 to 480 and the list goes on.

No offense but raw performance isn't the only reason I bought my 680s and only 25 to 30 percent is a very solid boost especially when it comes it at a much lower power cost. And honestly that 30% number is only an average, in a number of cases it's well over that. I run 3D Surround and all I know is that I'm getting much better playability in games and able run at higher settings and still noticeably higher framerates.

I'm very happy with the 680s, it's one of the best upgrades I've ever done coming from 3 580s.
 
Actually Charlie is flat out saying that there is a defect in the 680. While I'm not inclined to believe him when it comes to anything about nVidia or his nVidia moles it is kind of hard to ignore that fact that nVidia doesn't seem to be able to produce the 680 fast enough and it does seem that others aren't having the problem as bad as nVidia, AMD seemed to have gotten its volume ramped up a bit faster, but we're only 5 weeks in with the 680 so well see how things go this month but if things don't improve with the 680s avilability then even I, as much I think Charlie lies as often as he tells the truth, might actually believe that maybe he might be correct in this instance.
 
No offense but raw performance isn't the only reason I bought my 680s and only 25 to 30 percent is a very solid boost especially when it comes it at a much lower power cost. And honestly that 30% number is only an average, in a number of cases it's well over that. I run 3D Surround and all I know is that I'm getting much better playability in games and able run at higher settings and still noticeably higher framerates.

I'm very happy with the 680s, it's one of the best upgrades I've ever done coming from 3 580s.

I would buy one too for power reasons if it were in my price range. Good reason since my 448 sends PSU in protection mode when I OC past 1000Mhz with 1.3V. But I have a cap of $300 period and require a significant boost to make upgrade. Big kepler will do it for me eventually. I can wait - I only game at 19x12 after all.
 
Good reason since my 448 sends PSU in protection mode when I OC past 1000Mhz with 1.3V.


Off topic but this is the first that I have of this PSU protection mode. Also how are you able to get 1.3v?
 
While I'm not inclined to believe him when it comes to anything about nVidia or his nVidia moles it is kind of hard to ignore that fact that nVidia doesn't seem to be able to produce the 680 fast enough and it does seem that others aren't having the problem as bad as nVidia, AMD seemed to have gotten its volume ramped up a bit faster,

Nvidia is using a different process than AMD though, so that could also be part of it. But Nvidia has had problems in the past making new designs manufacturable (for lack of a better word) so I wouldn't be too surprised if Nvidia was at fault here.
 
Off topic but this is the first that I have of this PSU protection mode. Also how are you able to get 1.3v?

Machine just shuts down, dead as a nail. You have to reprogram bios with NiBiTor - I can show you how through pm if you want.

I really need a new PSU but am having problem finding high wattage quality which will fit.

Rad is 240mm+ long and sits up top inside the old wave master and only 5.9" PSU will fit - problem is anything high watts in 6.3" these days.:(
 
Last edited:
No they should bring out real Kelper instead of hopped up GF104. Something worthy of upgrade. No offense GTX 680 owners but you only got 25-30% boost compared to over 60% from 280 to 480 and the list goes on.

If you're going to do 280 to 480, make sure you're thinking 480 to 680 then, as we all know the 580 was just Fermi fixed.

err...no offense taken :D, but I'd prefer 30-35% from 580 to 680 (measured actually).
 
Machine just shuts down, dead as a nail. You have to reprogram bios with NiBiTor - I can show you how through pm if you want.

After I installed 301.23 beta driver my card started to down clock on benchmarks when slightly over 1000mhz. The thing is that my card runs cool and before the driver I didn't have any problems. I have done a clean uninstall and tried different drivers but the problem keeps coming back. Not sure if it is a problem at the BIOS level of the video card or if my PSU is going bad. I really don't think 448 is going bad.

When my card down clocks it doesn't shut the system off. It just pauses for a fraction of a second during game or benchmark and then the card is down clocked.


Also could be there is a hidden registry that the 301.23 driver effected that I need to delete. I really want to figure out how to fix this.

ZoNe pm me with more information about increasing the voltage limit to 1.3v. I already have up to 1.213v which has cleared some benchmarks with gpu at 1.1GHZ. My temps are perfectly safe so I think I could possibly use 1.3v to push card further.

Big problem is the down clocking issue.
 
Actually Charlie is flat out saying that there is a defect in the 680. While I'm not inclined to believe him when it comes to anything about nVidia or his nVidia moles it is kind of hard to ignore that fact that nVidia doesn't seem to be able to produce the 680 fast enough and it does seem that others aren't having the problem as bad as nVidia, AMD seemed to have gotten its volume ramped up a bit faster, but we're only 5 weeks in with the 680 so well see how things go this month but if things don't improve with the 680s avilability then even I, as much I think Charlie lies as often as he tells the truth, might actually believe that maybe he might be correct in this instance.

Low production volume doesn't mean there's an inherent defect in Kepler. That would imply that all current cards were faulty, and it seems like everyone who has them (including yourself) is doing just fine with them.

EDIT: Just re-skimmed the article and what he is saying is that there is a (possible) defect in the manufacturing of Kepler, not the design or chips themselves. Just something that is causing low yields supposedly.
 
After I installed 301.23 beta driver my card started to down clock on benchmarks when slightly over 1000mhz. The thing is that my card runs cool and before the driver I didn't have any problems. I have done a clean uninstall and tried different drivers but the problem keeps coming back. Not sure if it is a problem at the BIOS level of the video card or if my PSU is going bad. I really don't think 448 is going bad.

When my card down clocks it doesn't shut the system off. It just pauses for a fraction of a second during game or benchmark and then the card is down clocked.


Also could be there is a hidden registry that the 301.23 driver effected that I need to delete. I really want to figure out how to fix this.

ZoNe pm me with more information about increasing the voltage limit to 1.3v. I already have up to 1.213v which has cleared some benchmarks with gpu at 1.1GHZ. My temps are perfectly safe so I think I could possibly use 1.3v to push card further.

Big problem is the down clocking issue.

Will do but U have to wait till I get home from shop. You can also fix throttle by reprogramming.

In meantime you know lots of HW maybe you know of a short high powered quality PSU? I only have 5.9" to work with as you can see.

IMAG0081.jpg

IMAG0080.jpg
 
ZoNe not sure what your budget is for PSU but I wouldn't skimp out on quality. I looked and looked and the only thing that I felt good about recommending is the Kiwnwin Platinum series. Not sure if the dimensions are correct on Newegg but if so it looks like some of the higher powered Kiwnwin Platinums might work in your case.

My favorite is Seasonic but they don't have anything as small as you need.

Good luck with PSU problem. I do think you need to upgrade your PSU. 500W doesn't seem enough for over clocking.
 
Low production volume doesn't mean there's an inherent defect in Kepler. That would imply that all current cards were faulty, and it seems like everyone who has them (including yourself) is doing just fine with them.

EDIT: Just re-skimmed the article and what he is saying is that there is a (possible) defect in the manufacturing of Kepler, not the design or chips themselves. Just something that is causing low yields supposedly.

Yeah, he's saying it's a process problem not a design flaw which is plausible.
 
One thing process guys like to complain about is that ASIC guys don't know how to design chips that are easily manufactured. Complicate that further with a foundry that doesn't have intimate communication with designers and it's very easy to have an issue such as the one Charlie alleges. ;)

I must say, it seems that AMD has a better track record with yields, probably due to their previous ownership of fabs.
 
Yeah, he's saying it's a process problem not a design flaw which is plausible.

That I can agree with. Yield issues are related to the process, not the chip's design itself so long as the chip was designed according to the specification laid out by the fabrication company.

But the GK104 design itself is not defective if its working perfectly fine.
 
So charlie is saying that only 10k gtx 680s were produced?

What the fuck? That isn't even remotely believable. Whoever is feeding this news to him is likely trolling him.
 
That I can agree with. Yield issues are related to the process, not the chip's design itself so long as the chip was designed according to the specification laid out by the fabrication company.

But the GK104 design itself is not defective if its working perfectly fine.

There are multiple designs. In this case, the allegations are that the logical design is fine, but the physical design is not.

I'm defining logical design as the gate layout and logic (schematic) and physical design as feature sizing (layout - size of transistors, vias, metal layers, ect).
 
The title is so misleading. All charlie is saying that the yields are low, the damn chip is not defective. They are not getting enough yields and that is all. You can call it defective if they are shipping the defective cards out. Yes I believe it because the supply has been terrible almost a month after release date, this only speaks volumes. So is charlie right? well I guess if you look around you can tell that it doesn't take a genius to figure out that Nvidia is having problems getting these things out. I will bet that the gtx 670's will be readily available. Almost ever frys store already has about 5-6 units in stock by msi.
 
So charlie is saying that only 10k gtx 680s were produced?

What the fuck? That isn't even remotely believable. Whoever is feeding this news to him is likely trolling him.

Not that hard to believe to be honest. When was the last time you saw the gtx 680 stay in stock for more than a day. 500 dollar card is not that hot the hundreds of them sell out all of the sudden.
 
Not that hard to believe to be honest. When was the last time you saw the gtx 680 stay in stock for more than a day. 500 dollar card is not that hot the hundreds of them sell out all of the sudden.

Well they certainly have plenty of them over in Europe. Maybe they accidentally shipped to many to the UK/EU and not enough to the US lol.

http://www.google.co.uk/search?hl=e...171618.0.171772.7.1.0.6.0.0.97.97.1.1.0...0.0.

http://www.google.de/#q=gtx+680&hl=....,cf.osb&fp=6c6b5a1bd0a2b957&biw=1680&bih=923

http://www.overclockers.co.uk/productlist.php?groupid=701&catid=1914&subid=2255
 
They also cost ~$740 in Europe.

Nope not in Germany. Here you can get a GTX680 starting from 479 Euro and up. The following list just contains the ones that are available. Behind each link are several online shops.

http://geizhals.de/?cat=gra16_512&bpmax=&asuch=&v=l&plz=&dist=&xf=132_2048~1439_GTX+680&sort=p

Availability in Germany is good. Whoever wants a GTX680 will get one here. It's not a problem at all.

The availability problem in the U.S. seems more like underestimating demand or misallocation. In the end demand for the GTX680 is way higher than for the AMD counterpart.
 
I want to respect the guy but I just can't. Whenever ATI does something good he acts like they are gods, but when Nvidia does something right, he can't admit they are right. There hasn't been a peep out of him about Nvidia since January, because they have been destroying ATI, but now that their may be a problem he jumps all over it.

He seems to have a hard time believing that the demand for the 680 is that high. This card is available everywhere from Asia to Europe to North America. I honestly doubt if they only had 10k chips, we could see such wide availability around the world on launch day.

Whats more, we're already seeing custom cards out there in the market. There's no way the board makers could make these custom cards if there was a severe shortage of the GPU.

Demand is just that high, and he just refuse to accept it.:p
 
Yeah, let's see...

Smaller die than 7970 = check.
Better scaling with more cards = check.
Smoother gameplay experience = check.
Faster gameplay experience = check.
Mouse cursor doesn't lag in the corner = check.
Diablo 3 supported with latest drivers = check.
In general better day one driver support = check.
Better drivers = check.

I guess the card is defective, should be recalled, etc, etc, etc, etc, etc. nVidia is bad!!!111! I really hope nVidia releases some sales numbers that Charlie can print out and eat :p.
 
I guess the card is defective, should be recalled, etc, etc, etc, etc, etc. nVidia is bad!!!111! I really hope nVidia releases some sales numbers that Charlie can print out and eat :p.

Just keep an eye out for an updated steam hardware survey... obviously not 100% accurate but it'll give us a quick peek when the 680 hits the list.
 
Just keep an eye out for an updated steam hardware survey... obviously not 100% accurate but it'll give us a quick peek when the 680 hits the list.

Maybe not, as it would have to be more than 0.5% to show up as its own category and not under "Other".

For instance, the 590 doesn't show up on the list and it's been out for a long time.
 
Well one thing is for sure if you look at the Steam hardware survey. More people use nVidia :p. Don't draw any conclusions about the sudden 'drop' in AMD near the end (like I almost did) -- there was some sort of bug affecting the pre-April numbers that Steam has explained at the top of the page.

It's funny that AMD marketing has this thread locked, because they claimed it was inaccurate. Then they go and post this thread. I do so love their hypocrisy.

AMD marketing at work people. More effective than their driver team since ... well ... it's been a while.
 
Yeah, frankly either the title needs to be edited on this or it needs to be closed entirely.
 
Well one thing is for sure if you look at the Steam hardware survey. More people use nVidia :p. Don't draw any conclusions about the sudden 'drop' in AMD near the end (like I almost did) -- there was some sort of bug affecting the pre-April numbers that Steam has explained at the top of the page.

Also, if you look at the trend there is this sharp jump in "other" GPUs, which I guess means they haven't catagorized some of the newer stuff yet? Or, like you said, there was some bug.
 
Also, if you look at the trend there is this sharp jump in "other" GPUs, which I guess means they haven't catagorized some of the newer stuff yet? Or, like you said, there was some bug.

In Steam's words...

There was a bug introduced into Steam's survey code several months ago that caused a bias toward older systems. Specifically, only systems that had run the survey prior to the introduction of the bug would be asked to run the survey again. This caused brand new systems to never run the survey. In March 2012, we caught the bug, causing the survey to be run on a large number of new computers, thus giving us a more accurate survey and causing some of the numbers to vary more than they normally would month-to-month. Some of the most interesting changes revealed by this correction were the increased OS share of Windows 7 (as Vista fell below XP), the rise of Intel as a graphics provider and the overall diversification of Steam worldwide (as seen in the increase of non-English language usage, particularly Russian).

Its possible the others are actually uncategorized AMD or nVidia drivers. You would think they could figure that out by looking at the installed driver.
 
In general better day one driver support = check.
Better drivers = check.

Bullshit, we're still on launch drivers well over a month later with bi-polar performance and a few odd issues.

Nvidia's launch support is horrendous and always has been. I remember when GTX460 launched the only available driver was some beta for more than two months and that was with an issue with X38/X48 chipsets.

Let me guess, the card runs fine with BF3 or some other popular game so all is well? :rolleyes:
 
Bullshit, we're still on launch drivers well over a month later with bi-polar performance and a few odd issues.

Technically, we're on a beta driver which is the second set to come out after the 680. But yeah, they are pretty slow to release WHQL.
 
Yeah, he did say all that garbage about PhysX being what would allow the 680 to beat the 7970, all of which turned out to be nonsense.

While that's true, it's also true that nVidia has been having more issues than the rest with respect to TSMC's 28nm. Whereas the others have reported supply issues due to lack of available wafers and not yields, nVidia has been having yield issues as well. Charlie, S/A or AT, it doesn't really matter who writes the article here as it's something that's been known for a couple of months now.

If it's true that they have to redo the mask then we're looking at 6 months+ added time. It could also be that TSMC overestimated their capability with respect to nVidia's architecture (think Llano + GloFo). Either way they're getting hammered more than the rest here.
 
Bullshit, we're still on launch drivers well over a month later with bi-polar performance and a few odd issues.

Nvidia's launch support is horrendous and always has been. I remember when GTX460 launched the only available driver was some beta for more than two months and that was with an issue with X38/X48 chipsets.

Let me guess, the card runs fine with BF3 or some other popular game so all is well? :rolleyes:

There's already a beta driver. If you only consider WHQL drivers, then AMD took more than a month. :p But lets no go down that path.
 
There's already a beta driver. If you only consider WHQL drivers, then AMD took more than a month. :p But lets no go down that path.

Read the release notes in that beta. They don't do anything for GTX680. They only add the driver optimizations that GTX680 already had that they held out on for the rest of their cards to make the performance at launch look a little better.
 
Nvidia is the only company having such extreme 28nm issues. Its not too hard to believe its their design causing it, and not TSMC.
 
Back
Top