Bitcoin Forum

I am curious if anyone has had a card from around june or before that have been minning non stop that have caused problems? I am specifically looking to see if anyone who games and mines bitcoins and if they have had artifiacting or more crashing or if the card died.

How OC'd are you with them, and are they adequately cooled?

I have had several cards i used to run alot which died or has been failing somehow
The issue is did overclock damage it or the constant stress from running almost 24/7
The problem is actually you never can say if they would have died with or without the constant stress
Overclock code N = non, L = Low ( 5 to 45 Mhz), M = (45 - 100 Mhz) and H= High (higher then 100 Mhz or more)

Nvidia 6800 GT, N, usage: Physx and gaming about 12 hours a day the hottest card i ever owned ( 123 C)
Nvidia 8800 GTX, L, usage : distributed computing 24/7, suddenly burned down
Nvidia 9600 GT, L, usage : physx + Distributed computing, from scratch gave artifacts slowly died
Nvidia 280 GTX, N, usage :physx _distributed computing, burned down
Nvidia 440 GT, L, bitcoin, died yesterday 4 months old
Ati 4830, H, usage: distributed computing, slowly ran every month lower clocks till sudden dead.
Ati 4890, N , usage : distributed computing, i think the enormous heat damaged it ( 112 C) still worked but xfx replaced

Another thing some cards at start i had made huge overclocks but in time i constant had to lower the overclocks since they give artifacts or crash at the previous settings
I had 4 x 4850 which ran at very high overclocks but slowly began to decay in speed at the end they hardly went above stock speeds.

I somehow find the ati cards however giving me much more stable cards if you keep them stock speeds.
Only one ati died out of 27, compared to nvidia 5 out of 8  

had a fan get noisy on an asus direct cu 6870 (1000/300) I bought in june 2011 running the fan at 85-90 % (this before using cgminer with its auto clocking and auto fan features) . oiled it and its fine now but Im not impressed. couldnt return it as I had already replaced the TIM with AC MX-2

been doing folding for years on various overclocked but well cooled cards. never killed a card but Ive killed some fans over the years. keep em under 75% if ya can.

I lost one after a power outage outisde of that no troubles

Yes (some faster than others, some not at all).  I run distributed computing (Boinc) projects before mining so have stressed cards longer than just June.  My graveyard covers quite a few.  Not running o/clock, and most are mem down clocked.

HD2600 - it was a cute card
HD4850 - several
HD4890 - nice card, ran hot
gt8800 - still going, but suspect
gt9800 - simple fail - not even artifacts
gtx280 - major artifacts on load, will not compute
HD5850 - tried the "oven bake" fix on this as a test, but used the fan in another card.
HD5850 - meh
HD5970 - XFX black edition, very nice paper weight
HD5970 - screens of death (blue, grey, green, black - I think the gtx280 was the only one with a pink/red screen of death)

Clearly, yes, chances of killing a card will go up considerably if you stress them 24/7, particularly if you overclock them and dont provide ample cooling.
Its simple really, running a card causes electromigration. Think of it as erosion inside the chip. Electromigration is worsened by:
 heat and voltage (and very much so!).

Its like running your car flat out. Its designed to run flat out, and it shouldnt break just because you do it, but chances of it breaking down do go up considerably if you do it 24/7. But its still only a chance. Was it saab that ran their cars for 200.000 Km flatout on a circuit some years ago as a publicity stunt? Dont remember.

Anyway, for some more anecdotal evidence, mining with old 8800GT, 2x5850 and a 4770. One of the 5850s died after a month.

I almost killed a 9600GT, my first mining card.  After a couple days of running it will start to artifact on its own not even mining, just showing a Windows desktop.

My 5830's have slowly lost overclocking abilities.  I used to be able to push over 1,000MHz, now I can barely hit 980MHz.  For a couple of weeks I was running super high voltage on them also.  24/7 mining + heat (especially if you crank voltage) will definitely kill a card.  The big boys that have large farms are no strangers to flat out smoking cards and having to replace them left and right.

technically you should breakeven before killing your card, but at today's difficulty/price ratio I'm not so sure anymore :(

got the cheapest 5770 i could find and it's been plugging away 24/7 since june- got a 5870 too but it's only active 50% of the time since I need it's RAAHW POWAH for producing stuff.... no problems at all for either of them at all.

here is a good analogy.

if you buy a car, put it on the highway and floor it, how long will it last?   myguess:  20,000 miles?

if you buy a car, put it on the highway and go 80 MPH.  how long will it last?   myguess:  180,000 miles? or more

You might be surprised.
http://www.youtube.com/watch?v=Ov8m8gJNeGA
For those not wanting to watch the boring video, saab put a couple of regular 5 door cars on a test circuit and floored them for 100000 Km. IIRC, they only changed tires and topped oil. This is 25 years ago with cars that frankly, where no good (Ive had one for a short while decades ago. To be fair it had 250.000 Km on it when I got it lol, so perhaps they were not that bad).

Anyway, the car analogy is somewhat flawed, as I think flooring a car non stop is probably less abusive than driving it in the city, doing short distances and lots of cold/warm heat cycles. Although some of that actually may apply to gpu's as well (the heat cycles).

Electromigration is the real killer. The longer you run the card, the hotter you run the card, and particularly the higher you set the voltage, the bigger the chance of a card failing due to electro migration. I cant think of a good car analogy, other perhaps than tires wearing out and eventually blowing, but more info here:
en.wikipedia.org/wiki/Electromigration (http://en.wikipedia.org/wiki/Electromigration)

edit: btw the last paragraph of that wiki article also explain why "baking" a card in an oven can sometimes fix it, if the electromigration occured in a solder joint. Heating it and melting the solder again, can actually fix it.

I got 2 5850's running at 86% fan and 86 degrees F.  Is that too hot?

No mate, that way too cool. you will wear your fans out in no time for nothing. but I kinda doubt anything other than chilled water cooling would keep the card at that temperature. Unless you mean 86 celcius, which I would consider too high.

Hummm... Double checked and it is Celcius but now the fan is at 76% and the temp is 60 C.  How hot is too hot?

Thats quite a drop. Something aint right there. Be VERY careful never to use 2 temperature monitoring apps at once, like GPU-Z, Everest, Trixx or afterburner. There is a bug in those cards that can trigger voltage to spike to 1.65V if two apps are monitoring  :
http://www.overclock.net/amd-ati/648462-hd-5870-random-voltage-bump-1-a.html

I killed a 5850 that way:
https://bitcointalk.org/index.php?topic=44495.0

It only worked for a few weeks after that happened.

Anyway, to answer the question. 60C is great. 70C is my personal pain limit. Anything over that is not a good idea to run 24/7. Its not like it will instantly fry at 85C, but its not good for longevity either.

also be sure to check VRM temperatures if you can. Not all cards support it, but if yours do, check GPU-Z. Try to keep those under 100C

I am reading this carefull and found indeed the same disturbing change in volts on my 5870 ... never knew nor even looked at it before.
Simply because i do never overvolt, i think you hit the hammer at the right spot with your answer
This might have killed the old ati card as well, all though that can never be proofed anymore.
Anyway i turned of 2 programs which where monitoring the hardware simultanious.
I decided only to keep afterburner running for keep an eye on my gpu's cause of the cpu gets too hot it gets turned of by the bios
On the temps i would say max for long term running is still about 90c, with some peaks near 100c anything above i consider dangerous if longer then a few minutes.
The cooler you can run them the longer they will last or die


brrr lol mine second cards do come over 100c since they are stuck to each other in the case

Amazing huh, by being careful and keeping a watchful eye on your hardware you end up killing it. And surprisingly its not well known.

As for your temps; Id really look in to an aftermarket cooler. Look at this asus slide:

http://event.asus.com/mb/5000hrs_VRM/images/5000hrsVRM_05.jpg

Ignore the absolute numbers, its for solid cap VRMS on a motherboard, which unlike with videocard VRMs, rarely blow out anyway. The point here is that that operating temperatures decreases life expectancy exponentially.

Unfortunately, I cant monitor VRM temps on my current cards, but the same logic holds true for the GPU itself. My 5850s dont break ~60C (on the default sensor but that also means around 70C on sensor #2 in GPU-Z). Thats with a standard cooler (albeit non reference). My 5870 with reference cooler should come in today, about the first thing Ill do is fit a VRM cooler (thermal right R5) and see if my accelero twin turbo is good enough.

Aftermarket coolers tend to be pricy if you are running a bitcoin farm, but you can almost always reuse them for a new card, even if some minor modding might be needed. So I bought a pile of battle-axe (http://www.techpowerup.com/reviews/Xigmatek/BattleAxe/) cooler second hand for 6 euro each, that will come in handy if I decide to expand further.

Not necessarily, but the short term risk of fan failure reaches 100% pretty fast if you run the cards 24/7 at full or very high fan speed.
So if you still have hardware in use for mining, there is a very high chance the fans will start failing one by one.

I didn't have *all* fans fail on my cards but a fair amount did.

IMHO, unless you have FREE or extremely cheap electricity, overvolting or overclocking is pointless because it will cost you so much more on your power bill.  Just pulling numbers out of my ass, think 25% more power for 5% more bitcoin.  Might as well run stock or overclock it to the point that's stock-like.

edit: not only the cost of electricity, but you end up saving a lot of money in cards not dying.

Overclocking actually increases efficiency at a given voltage. Power consumption of the card scales at most linearly with clock, meaning the entire rig's (including CPU, ram, motherboard etc) power scales less than linearly with the gpu clock. Performance does scale linearly, so MH/W improves as you overclock the gpu. Overvolting OTOH is of course not going to increase power efficiency, its going to reduce it quite dramatically.

Because this has popped up in my un-read section, I can add another 5970 to my dead pile, but I did also swap a fan between my XFX 5970 into the sapphire card.

I've had problems with 2 of my 9 cards which went live April/May 2011 time.
Card 1 died on me around October. The fans would spin up on power on but no post or display, upon further inspection it was apparent that a single cable from my pcie plug had come loose and fell out. Not sure if that was the cause or just a coincidence.
Card 2's fan bearings seized on me just before Christmas. Bought an aftermarket cooler as an rma would have left me with an alternate model.

Both cards were overclocked to just before the point of crashing, fan speed a constant 85% and temps were always under 70C, even in the hight of summer. They were made by XFX to which I am trying to avoid in the future  :D

I've been mining since June with with about 10 GPUs. Most of them 5870s, with some assorted 5850s, 5770s and a 5970 thrown in themix.

EVERY. SINGLE.  ONE of them now has problems. Some of them have outright failed, others just artifact and are still good for mining, but not really as a display card anymore. I can't discount 10/10 cards as mere coincidence. So yes, they definitely do kill your cards.

What clock, voltage, fans, and temp.  I have been mining w/ 15 5970 for much longer than that and they are all solid.  I did have to replace 2 fans so far and likely will need to replace them all eventually but that is because they are Chinese junk.  Then again I don't abuse them with high temps, voltages, and clocks.

Mostly had problems with the fans, only one fried chip on a XFX 5970

For me it's 0/7 GPUs. All in fine working order, the oldest of them running since summer. Undervolting is crucial IMO.

They've not been idling, either. The ASUS hd6950 DCII 1GB (which I detest for its non-standard VRM, BTW) has been running since early September at 942 MHz. This is the max stable setting for this particular card using any recent AMD driver and 2.5 SDK.  At 945 it will lock up in minutes.

Scoring a perfect 10/10 failured cards, surely you must have done something wrong?  :o
May I ask what temps and fan speeds you've been running those cards at?

 I don't think videocard can die if everything tuned properly, otherwise it can die even from playing games with high AA @ stock frequencies!

 Look at my radeon 5770 stats:

850(stock)/300:
temp1: 69.5
temp2: 74
temp3: 73
fan speed: 51%
VDDC:1.125 v
 191.31 mhash/s

850/1200 (fully stock):
temp1: 74
temp2: 80.5
temp3: 78
fan speed: 56%
VDDC: 1.125 v
196.59 mhash/s

960/300:
temp1: 73
temp2: 78
temp3: 76.5
fan: 55%
VDDC: 1.125 v
221.10 mhash/s

 Thats funny, but OCed it's even a bit colder than stock and with much higher mhash/s! So, if it will die, then it will die with stock freq. earlier than with 960/300! And difference in temp and fan speed with 850/300 is very tiny.

 UPD with undervolted card.
960/300/1.005v:
temp1: 67
temp2: 71
temp3: 70
fan: 50%
221.39 mhash/s

Been running the 5870's at 900 Mhz, underclocking the memory to 1000. The 5970 I left at stock because I wanted it to last awhile (lasted all of 2.5 months until 1 GPU just outright failed lol). Stock voltages.

Most of my cards live at around 80 celcius, with some of them getting to be in their own cases living at 70-75 celcius instead.

Two of them have the weirdest problem: they will be fine under load, but when you let them idle, then they display erratic screens. lol.

For those of you above who said that you have had 0 fail, have you tested them as display monitors? Where you hook them up to 2 screens? I ask because for some of the cards, they are fine on 1 screen but if you ask them to display two then they start artifacting, which to me is still a failure.

And its not like videocards that are not used for mining ever die  ???

Its really simple, any silicon chip thats used will eventually die. Its a given (http://en.wikipedia.org/wiki/Electromigration). The only question is how long it will take. It could be centuries, it could be days. But stressing these chips will reduce their lifespan, particularly when overvolting and overheating. If you are lucky, you may never notice, but it doesnt change the facts.

I know. I mean it should work at least 2-3 years with 24/7 mode and with temp <70C. Overclocking + undervolting will kill a card faster if temp is in safe spot (<70C), than stock freq/volt with >80C?

undervolting should make the card last longer, because lower voltage = smaller electromigration

Obviously not. Both current and temperature are key exacerbating factors of electromigration. Clockspeed afaik, is not. But its a false dilemma, because a higher clockspeed will result in higher temperature and require higher voltage, regardless of what your testing shows (hint: there are several temperature sensors in the GPU, you are measuring only one, quite possibly GpuIO which is the memory controller. Try checking the shader sensor).

Temp1-3 are different sensors, no? I don't see any others with gpu-z.

INdeed, you posted all 3. Still its worth pointing out that your 850/300 temps are unsurprisingly lower than your 960/300 temps despite lower fan speeds, so my point stands. Its also a fair point to make that gaming cards are not designed to run what AMD would consider a thermal virus 24/7. Professional compute cards from AMD and nVidia (firepro and quadro) are mostly based on the chips as the gaming cards, but they generally run at lower clocks and voltages. These cards are designed to run 24/7 at max load.

Well, but temp of 960/300/1.005v is lower than 850/300/1.125v. ;)

Seasonic and Corsair PSUs. I also don't run them at > 65% of their full advertised capacity, so it shouldn't be PSUs. Temps might be it, but I noticed that with air flowing on them vs air not flowing on them from 120 mm fans, the temp only dropped about 1-2 degrees celcius, so *shrug*. Don't know how I could have made them any cooler without running caseless.

I ruined a 5970's by setting it to start mining on system startup. When the pc starts all the fans go to 100% and the pc is very unstable for a few seconds. Topped with mining seemed to screw up one of the gpus. I finally got it mining again, but it can't be used for anything else now. Luckily that is the only thing I bought it for.

Definitely have this issue.

I found that artifacting happens around the time I start changing clocks and applying a load to the cards. It mines fine, and it is NOT a clock I would normally set for gaming. If I reboot and use stock or "gamer" type overclocks and run furmark, it runs fine no artifacts.

This.

I've got a Radeon HD6870 (Powercolor) and It's only been running for about two weeks, I haven't noticed anything wrong with it at all, but it does get a bit hot for my liking, GPU-Z reports:
                     Temp 1: 68.5C
                     Temp 2: 79.5C
                     Temp 3: 77.0C
                     Voltage: 1.17V
Is this normal? No overclocking, I just want to know if anything will happen due to these temperatures, as I am an avid gamer on this PC.
GPU Core lock is at 900MHz and the Memory clock is at 1050mhz, getting 250-270MH/s

Underclock your memory, as much as you can. If wont hurt performance (on the contrary actually) and it will lower temps a little bit. Even so those temps are uncomfortably high IMO.  Try to improve air flow in your case or speed up your fan. If your card supports it, you can also try reducing voltage.

How low would you recommend, stock is 1050MHz
Also, improving airflow is out of my range since I've only got one card and I've got a couple 1300RPM fans pointed right at it, 95 CFM each, they move a lot of air.
My fan has been spinning at 100%. Powercolor is very bad at cooling. I would buy a new heatsink for it, but it voids the warranty, and this card may be one of the defective ones they sent out this past couple months.
Edit: I downclocked the memory to 900Mhz and noticed a significant temperature drop. GPU-Z reports:
Temp 1: 61C
Temp 2: 71C
Temp 3: 68C
How much lower can I go without anything bad happening?

Mining doesn't need a fast memory clock. I run 5970s at 150 Mhz.  The ban news is without a bios flash (risky) 6000 series only allow mem clock that is 150 Mhz lower than core.  So if you are running 900 Mhz core you can try 750 Mhz memclock.   Remember if you later raise the core clock you need to raise the memclock.   Nothing "bad" will happen if you try to set it too low.  The card will simply ignore the request and run it at stock.

Looks like the 7000 series don't have this "glitch". Only the 6000 series does.

I have flashed my bios to the latest stable version already for my Phenom II X6 processor.
Does this mean I can go lower than 150MHz lower?

What the frak?
You're suppposed to flash your GPU bios, not the one on your mobo...
Try the search button, this subject was discussed many a time.

Also, while I hate to discourage you, your having made this mistake tells me you should probably not flash anything until you have done your research.
RTFM first, understand it, then flash. You're risking your hardware otherwise.

Damn, that's sucks.

LOLZ.

D&T PSA:
Just to be clear.  If you don't know what you are doing DON'T flash your GPU bios.  It is a good way to end up with a $300 paperweight.  It is possible to put a custom bios on the card to make it do whatever you want.  That is the only way around the 150 Mhz mem limit on 6000 series.  It is somewhat risky, very easy to kill your card if you do something stupid and is 100% not supported by any warranty. 

End D&T PSA

I flashed my BIOS 3 months back because my processor didn't work. I'm sorry if that's not what he meant, and I'm not going to flash my GPUs BIOS anyways.

Edit: Wording