Modern Graphics Card failure rates

[thilanliyan]

http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

 

[zerogear]

You would think so, but apparently people don't care. Look at XBox 360, people buying 3-4 XBoxes that keep on failing.

 

[thilanliyan]

Originally posted by: zerogear
You would think so, but apparently people don't care. Look at XBox 360, people buying 3-4 XBoxes that keep on failing.

That's true but they sorta have to right? Otherwise their game collection would go to waste...with video cards you don't HAVE to stick to one vendor thankfully.

If I had bad luck with several cards from one vendor...I'd definitely try the other to see if I have better luck with them.

 

[taltamir]

Originally posted by: thilan29
http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

 

[LOUISSSSS]

Originally posted by: taltamir

Originally posted by: thilan29
http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

not always a free upgrade

and you're forgetting the shipping and packaging cost of ~$10

and you're also forgetting the 2 week downtime of a pc without a high end video card or no video card at all if you dont have a spare.

 

[LOUISSSSS]

gtx280 - 10% failure rate. thats pretty bad...

 

[zagood]

Originally posted by: taltamir
XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

These vendors aren't even part of the survey. I wonder if the article's failure rates still hold with their parts.

 

[Soulkeeper]

it don't surprise me, as i had to return a gtx260 a few days ago
I think they've just gotten so complex/big/hot it's hard to maintain high yeilds

 

[Leyawiin]

Originally posted by: LOUISSSSS

Originally posted by: taltamir

Originally posted by: thilan29
http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

not always a free upgrade

and you're forgetting the shipping and packaging cost of ~$10

and you're also forgetting the 2 week downtime of a pc without a high end video card or no video card at all if you dont have a spare.

And the fact you're oftentimes getting someone's "refurbished" problem or beat-to-death step-up turn in.

 

[Cookie Monster]

I think i know what contributed to the mass GTX280s failing. Upon its initial release, there was a batch of cards that just overheated without warning. This problem has been fixed as of today but back then, alot of these cards were "overheating" i.e led to failures (There is a massive sticky in nvnews forums). Im not sure if this anomaly was publicly addressed. Im guessing it had to do with nVIDIA limping on the verification process of the GT200 cores, loosening on some testing/simulation procedures to get as many "good" cores for their GTX series. Since yields weren't so great back then and they required this product on the shelves (GT200 was delayed for about ~7 months), they needed as many of these to meet the demand, not to mention to compete against the competition.

 

[MarcVenice]

Guys, take these numbers with a grain of salt. The samples aren't actually that big. It's possible that 1 card failed, and adds 1 or even 2% to the failure rates. It's also possible that a card has been send back by a unhappy customer, but that the card itself was working just fine.

 

[BFG10K]

I'm surprised to see Gigabyte doing so badly.

 

[zerogear]

Originally posted by: Cookie Monster
I think i know what contributed to the mass GTX280s failing. Upon its initial release, there was a batch of cards that just overheated without warning. This problem has been fixed as of today but back then, alot of these cards were "overheating" i.e led to failures (There is a massive sticky in nvnews forums). Im not sure if this anomaly was publicly addressed. Im guessing it had to do with nVIDIA limping on the verification process of the GT200 cores, loosening on some testing/simulation procedures to get as many "good" cores for their GTX series. Since yields weren't so great back then and they required this product on the shelves (GT200 was delayed for about ~7 months), they needed as many of these to meet the demand, not to mention to compete against the competition.

Actually, most of those were loose TIM on the chip/heatsink, which a lot of people could've fixed by just taking the heatsink off and reapplying. Which is not to say is acceptable, but it wasn't a hardware failure on their part, but a manufacturing QA problem

 

[Keysplayr]

Originally posted by: LOUISSSSS

Originally posted by: taltamir

Originally posted by: thilan29
http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

not always a free upgrade

and you're forgetting the shipping and packaging cost of ~$10

and you're also forgetting the 2 week downtime of a pc without a high end video card or no video card at all if you dont have a spare.

That's why every computer enthusiast should have a spare PCI-e card laying around for just such an occasion. I do. 8400GS. Yeah, I won't be able to game until RMA is fulfilled, but that might be a plus considering how much time it wastes.

 

[BFG10K]

Originally posted by: keysplayr2003

That's why every computer enthusiast should have a spare PCI-e card laying around for just such an occasion.

Or a GMA.

 

[thilanliyan]

Originally posted by: MarcVenice
Guys, take these numbers with a grain of salt. The samples aren't actually that big. It's possible that 1 card failed, and adds 1 or even 2% to the failure rates. It's also possible that a card has been send back by a unhappy customer, but that the card itself was working just fine.

Does it show the sample size? It says the that data was collected from August 2008 to March 2009...that seems like a fairly long time. It is missing some large vendors though.

 

[OCGuy]

I can definatly believe the 280 being that high at launch (there were thermal-paste problems I believe. I think even a mod here got a bad one).

I would have liked to see some of the US big guys on that list.

 

[Lonyo]

Originally posted by: thilan29

Originally posted by: MarcVenice
Guys, take these numbers with a grain of salt. The samples aren't actually that big. It's possible that 1 card failed, and adds 1 or even 2% to the failure rates. It's also possible that a card has been send back by a unhappy customer, but that the card itself was working just fine.

Does it show the sample size? It says the that data was collected from August 2008 to March 2009...that seems like a fairly long time. It is missing some large vendors though.

The comments at the bottom say minimum sample size was 500, but the http://www.hardware.fr/news/10...cartes-graphiques.html site might suggest sample size of minimum 100 for the card vs card test.
Brand failure rates should certainly be fairly representative though.

 

[Cookie Monster]

Originally posted by: zerogear
Actually, most of those were loose TIM on the chip/heatsink, which a lot of people could've fixed by just taking the heatsink off and reapplying. Which is not to say is acceptable, but it wasn't a hardware failure on their part, but a manufacturing QA problem

See thats what I thought initially, but some cards even with reapplied TIM still had this problem. Take a skim through the nvnews thread

Here are just a few quotes that massive thread to prove my point.

Originally posted by: pingu2
my gtx280 is also running hot (or says it is), I have applied arctic silver etc but its not made any difference.. even with fan running at 100% it reaches 105c in vantage test 2 - however I am not convinced it is actually reaching that temp - it doesnt feel that hot, and I get no artifacts etc.. it throttles back when it says its at 105c and then runs fine at 104/105 indefinately (by alternating between full/throttled speed)

Originally posted by: Ravynmagi
I went ahead and replaced the thermal paste on the GTX 280 and I've reseated the heatsink at least four times as I've re-inspected it to make sure the paste is showing sign of full contact with the heatsink. I'm pretty positive that the heatsink and ambient conditions aren't an issue, at least in my situation. And there clearly is a big difference between the temps a few of us are getting and the temps others and reviewers were reporting. It does seem plausible that most people are getting 85 to 90C max load temps. While some of us seem to have cards going over 100C.

Originally posted by: Malichite
The overheating problem can be repeated in under 1 minute running Ozone3d FurMark from a cold boot, so ambient temps are not an issue. Normal GTX chips seem to hold in the low 70s C to mid 80s C, while the anomalous chips rapidly rise to 100+C and either throttle or shutdown. Several members have also tried to reseat the heatsink using a fresh application of AS5, but has no positive effect. Given that the overheat is so repeatable at low ambient temps, it is clearly another problem. Earlier in this thread MikeC said he also noticed something similiar with his nVidia reference card, but had no problems with his eVGA review card so it seems like a few bad chips slipped through quality control.

 

[zerogear]

Originally posted by: Cookie Monster

Originally posted by: zerogear
Actually, most of those were loose TIM on the chip/heatsink, which a lot of people could've fixed by just taking the heatsink off and reapplying. Which is not to say is acceptable, but it wasn't a hardware failure on their part, but a manufacturing QA problem

See thats what I thought initially, but some cards even with reapplied TIM still had this problem. Take a skim through the nvnews thread

Here are just a few quotes that massive thread to prove my point.

Originally posted by: pingu2
my gtx280 is also running hot (or says it is), I have applied arctic silver etc but its not made any difference.. even with fan running at 100% it reaches 105c in vantage test 2 - however I am not convinced it is actually reaching that temp - it doesnt feel that hot, and I get no artifacts etc.. it throttles back when it says its at 105c and then runs fine at 104/105 indefinately (by alternating between full/throttled speed)

Originally posted by: Ravynmagi
I went ahead and replaced the thermal paste on the GTX 280 and I've reseated the heatsink at least four times as I've re-inspected it to make sure the paste is showing sign of full contact with the heatsink. I'm pretty positive that the heatsink and ambient conditions aren't an issue, at least in my situation. And there clearly is a big difference between the temps a few of us are getting and the temps others and reviewers were reporting. It does seem plausible that most people are getting 85 to 90C max load temps. While some of us seem to have cards going over 100C.

Originally posted by: Malichite
The overheating problem can be repeated in under 1 minute running Ozone3d FurMark from a cold boot, so ambient temps are not an issue. Normal GTX chips seem to hold in the low 70s C to mid 80s C, while the anomalous chips rapidly rise to 100+C and either throttle or shutdown. Several members have also tried to reseat the heatsink using a fresh application of AS5, but has no positive effect. Given that the overheat is so repeatable at low ambient temps, it is clearly another problem. Earlier in this thread MikeC said he also noticed something similiar with his nVidia reference card, but had no problems with his eVGA review card so it seems like a few bad chips slipped through quality control.

Unfortunately, the area between heatsink and the actual chip is quite large, and many times even when they apply a new thermal paste, they don't put enough so that the TIM in between the chip and heatsink meet. (Arctic Silver does a pretty bad job at this since you need a fairly thick layer to have the chip and heatsink meet). I've taken apart quite a few of my friends GTX 280s to look at the new layer they applied, and the tim barely touch the heatsink. I applyed a thermal pad to between the chip, and everything went back to 60-70C idle, where before it around 95C idle

 

[dguy6789]

Originally posted by: taltamir

Originally posted by: thilan29
http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

I had an eVGA 7900GT that they refused to repair or replace when it died.

 

[Cookie Monster]

Originally posted by: zerogear
Unfortunately, the area between heatsink and the actual chip is quite large, and many times even when they apply a new thermal paste, they don't put enough so that the TIM in between the chip and heatsink meet. (Arctic Silver does a pretty bad job at this since you need a fairly thick layer to have the chip and heatsink meet). I've taken apart quite a few of my friends GTX 280s to look at the new layer they applied, and the tim barely touch the heatsink. I applyed a thermal pad to between the chip, and everything went back to 60-70C idle, where before it around 95C idle

Im not denying the fact that re-applying the TIM successfully to make full contact with the stock HS could fix the "overheating" issues prevalent in the early days of the cards retail life, but fact of the matter is that some still had this issue even if they've applied in 4 times in row (refer to one of those quotes). The people who visit hardware forum sites aren't exactly your average joes either so questioning their ability to reapply the TIM holds not so much value.

Its very hard to think that nVIDIA limped on the QA process of their cards. The initial versions of these cards had complex PCBs, a big HSF and what not (i.e being very expensive), and to limp on the their QA procedures I find it hard to believe especially when they needed as many as possible. What makes more sense IMO is the actual earlier revision of the chip itself (or a certain batch). A 576mm^2 chip with 1.4billion transistors on the old 65nm process isn't a easy feat when one looks at it from a manufacturing point of view, especially when these are required to meet the demands of thousands.

That being said, its all a theory since nothing much has been revealed. Luckily this anomaly was only around for a short duration.

 

[bryanW1995]

Originally posted by: keysplayr2003

Originally posted by: LOUISSSSS

Originally posted by: taltamir

Originally posted by: thilan29
http://gpucafe.com/2009/03/mod...cs-card-failure-rates/

Interesting but probably inconclusive. I wonder what the sample size was? Thankfully I've been lucky...no failures with any of the various nV and ATI cards I've owned.

IF that 10% failure rate for 280s and 3870x2s is true...that's unacceptable IMO.

XFX, BFG, eVGA...
Lifetime warranty means that a failed card = free upgrade... I don't see a problem here.

not always a free upgrade

and you're forgetting the shipping and packaging cost of ~$10

and you're also forgetting the 2 week downtime of a pc without a high end video card or no video card at all if you dont have a spare.

That's why every computer enthusiast should have a spare PCI-e card laying around for just such an occasion. I do. 8400GS. Yeah, I won't be able to game until RMA is fulfilled, but that might be a plus considering how much time it wastes.

I snagged one of those 9600 gso's for $37 AR last year...I can still game ok with that when necessary... life's too short to spend 2 weeks with no gaming at all!!

 

[Zap]

Originally posted by: keysplayr2003
That's why every computer enthusiast should have a spare PCI-e card laying around for just such an occasion. I do. 8400GS. Yeah, I won't be able to game until RMA is fulfilled, but that might be a plus considering how much time it wastes.

Who says you can't game on an 8400 GS? Sure, you'll have to tank your graphics settings and still get low framerates, but it should be capable of playing most games. FWIW I once sold the video card out of my main rig (x800XL) and had to make do with a GeForce 6200 for around 3-4 months. I just turned stuff down and kept gaming. Of course at the time most of what I played was WolfET and WoW, but gaming is still gaming.

Originally posted by: zerogear
Unfortunately, the area between heatsink and the actual chip is quite large, and many times even when they apply a new thermal paste, they don't put enough so that the TIM in between the chip and heatsink meet. (Arctic Silver does a pretty bad job at this since you need a fairly thick layer to have the chip and heatsink meet). I've taken apart quite a few of my friends GTX 280s to look at the new layer they applied, and the tim barely touch the heatsink. I applyed a thermal pad to between the chip, and everything went back to 60-70C idle, where before it around 95C idle

That's a good point. Arctic Silver seems to be made for really thin applications. What is needed is a compound that works well in thicker applications, such as Shin Etsu stuff.

Originally posted by: dguy6789
I had an eVGA 7900GT that they refused to repair or replace when it died.

Did they give you a reason?

 

[Griswold]

Originally posted by: Lonyo

Originally posted by: thilan29

Originally posted by: MarcVenice
Guys, take these numbers with a grain of salt. The samples aren't actually that big. It's possible that 1 card failed, and adds 1 or even 2% to the failure rates. It's also possible that a card has been send back by a unhappy customer, but that the card itself was working just fine.

Does it show the sample size? It says the that data was collected from August 2008 to March 2009...that seems like a fairly long time. It is missing some large vendors though.

The comments at the bottom say minimum sample size was 500, but the http://www.hardware.fr/news/10...cartes-graphiques.html site might suggest sample size of minimum 100 for the card vs card test.
Brand failure rates should certainly be fairly representative though.

If my little french knowledge doesnt disappoint me again, it states that the minimum sample size is 500 for all but two models. And these two models (GV-RX26P5H & ENGTX280/HTDP/1G) come with a greater failure rates than 10% but only a sample size greater than 100 pieces.

Reliable figures I would say.