Bitcoin Forum

I have a bunch of Sapphire 5850 that that began failing on their own after 1 year and a few days.

At this rate I fear they'll all die on me a few month after warranty.

I know this is illegal, but given the low ethics of this company I'd feel no shame.

I was thinking on swapping their serial # stickers with newer cards. (2 week newer  :( )

Do the cards have serial numbers inside they will tap into that could void this RMA ?

 Ps : This sticker seems like a pain to remove.

Yep, the sticker are "tamper resistant" so it should be hard to remove the sticker without leaving traces that the stickers have been swapped.

And the GPU has a serial number, and if suspect, I think they would compare their records of the GPU serial with the card serial.
And what do you do when the newer cards fail after those 2 weeks? Then you don't have any sticker to swap with?
(you cannot use the "replacement cards" sticker, since these stickers have then expired*)

* Think of you have 2 weeks left of warranty. When replacing, your warranty is NOT renewed, rather they set the serial number to have 2 week left of warranty. When the newer cards fail, those 2 weeks are out. When the replaced cards fail after 1 year from replacing them, your warranty is like 2 years old and expired.

The cards propably log runtime too, so they can see in a log how much time the card have been "on". Propably the card logs full-load time too. So then they can see that the cards have ben run for longer than manufacturer date and that would raise some alarms.

Propably, the Sapphires 5850's are tuned to work flawlessly under full load until end of warranty (1yr) + a few days, to keep the parts cost down. Half load = 2 yr + a bunch of days.
25% load = 4 yr + 1 month
and so on. Since they don't expect people to game under full load 24/7 (which you would basically "do" if you mine with the cards, from the card's point of view), they don't expect the card to fail after 1 year.


If they find out what they do, you would be charged with fraud. You cannot expect the card to work after 1 year (when the warranty expired) so just pull up your wallet and buy a new card!

Swapping stickers isnt a good idea ...

Wouldn't planned obsolescence be a fraud too ? 
(cards were running at 1.088v and ~70c)

Ps : They aren't risking much since these card aren't sold anymore.

Was there no forethought when you bought the cards?  If you bought in on pretty much an AmEx card and some Visa/MCs the card will extended the warranty out a year.  If you knew you were going to min with the cards and didn't want the risk you could have also bought an extended warranty from somebody like squaretrade.  If taking the time and energy to carefully swap out serial number stickers is not worth the $20 to sell your ethics (the price of a squaretrade warranty) then you probably shouldn't be posting on this board.

I lie all the time when its a matter of convenience (like not bothering to tell Frys that a "brand new" video card is dead - good lord anybody knows you would have to stand in line for 1 hour only to see them put the card back onto the shelves) but not if it means scamming a company.  To each their own  ::)

Not everywhere warranty is 1 year

Here it is 2 years for example

Get Cleverer, and buy cards with longer warranties next time.

It's not planned obsolescence. We're basically stress testing these cards 24-7 for a year on end. Build out of warranty failures into your business plan, purchase extended warranties, etc. Your rationalization is ridiculous.

+1.

I consider it sort of lucky that the manufacturers still warranty the cards at all, considering the torturous non-typical usage we put them through.  I would imagine there's some sort of "normal use" clause in their warranty contract that they could utilize if they wanted to.

What is abnormal playing video games 24/365 ?

Seriously, I don't think because it's used mostly by gamers that our use of it is abnormal.  ATI state mining as a feature of their card on their website.

It would never come to my mind that an intel CPU would not perform it's job for centuries, how is this so different ?

I am personally monitoring my cards to make sure temps are never out of range, fan is ok, and heatsink is clean, ~65c in open air.   I think these are better working condition than a gamer with a poor ventilated case that leave heatsink get clogged with dust .

Getting serial # from the card is pretty trivial I would be surprised if they didn't pull the internal  serial # and compare it to the sticker when processing an RMA.  I would imagine the OP isn't the first to think of this.  

Then again maybe they are lazy and don't check.  For the OP it is essentially a wager where you are wagering a good warranty (50 weeks remaining) against the hope that you will get this card repaired (instead of the second one).  
Seems like a foolish bet to me.

No semiconductor will last centuries.  It is unlikely they even would last a decade of continual use.  As chips get smaller their lifespan will decrease.

http://en.wikipedia.org/wiki/Electromigration

Ok, you got me there (with the bolded bit).

A CPU is a much simpler device.  I would bet that 99% of the time, the main GPU chip itself is not what fails - failure is usually due to some other component on the GPU board, like capacitors, microfractures in traces, etc.  Likewise, if a CPU was integrated in a motherboard, you'd find that they fail much more often, but 99% of those failures would be motherboards, not the CPU itself.

http://www.cse.psu.edu/~yuanxie/Papers/DAC06.pdf

 :o OMG I didn't knew MTTF could be that low.

It's hard to get clear data on ASIC's MTTF.

I have a AMD K6 that begs to differ.

Your K6 has been running at 100% load continuously for 10 years?  I think not.

Even if it had it had a gate size of 350nm.  A modern CPU/GPU has a gate size of 28nm to 40nm.  The smaller the gates become the less material that can be eroded from electromigration before gate failure occurs.

Still I said "unlikely".  Every chip suffers from electromigration from the moment current is applied to the moment it fails.  There is some variance in the failure rates but every chip will fail. 

Now the 100btc question : What is the expected effect of electromigration on lifetime of 40 and 28nm parts from ATI ?

I do this all the time ... that's why I am a scammer :P

This behaviour is driving the GPU prices up for everyone >:(

Also, don't forget to OC the card to 1100 MHz clock then when it dies blame it was DOA or it broke by itself for extra scam points !

You're an idiot.  Overclocking doesn't break cards.

No, it's only been, I dont know, 15 years since I've bought it. You have to remember old chips like these dont have power-gating like newer chips nowadays, so yes, idling is the same thing as loading it up, as far as electromigration goes.

No one has studied the effects of electromigration on 40 nm chips for a decade because they haven't been around for a decade! But if you run them on stock clocks at undervolted settings, my money is on that they will last a decade.

Also, I find it a bit preposterous to consider that electromigration is the dominant factor in graphics card failure over time. I think its more likely one of the those capacitors leak or the MOSFETs blows.

Electromigration is mostly a factor of the fabrication process. There are professional devices that are built on the same exact process (and sometimes even using the same chips) that are required to operate for a decade.

I think it is a bit preposterous that you believe anyone said it was the dominant factor.  Hint: nobody did.

And yet you link to an Electromigration wikipedia page in answer to Transisto's question that a graphics card would not perform its job for centuries?

"centuries" =/= "the dominant factor in graphics card failure"

Obvious troll is obvious.

Why list one one of the meaningless, complex, and rare failure modes when there's a much easier and obvious one for a novice to failure analysis to grasp?

I think you're the one trolling here.

He asked a question I gave him an answer. 
Transisto didn't ask what is the most likely cause of failure.  He asked why won't a microprocessor last a century.

Not every post in the thread has to directly relate to the OP.  Case in point your idiotic past 3 posts.  Feel free to have the last word.

And you chose one of the more difficult explanations as a goal to make sure he could not understand what you were saying?

Actually, you didnt even explain it, you just linked to it. Way to contribute to a discussion, idiot.

That's why overclocking does not void your warranty. Idioto !

I've had a 1055T overvolted to 1.5v and loaded for pretty much all of its couple year lifetime, and no problems yet...  Keep it cool and it's hard to fry.  The problem is probably insufficient cooling and cheap PCBs/MOSFETs/chokes.