What is killing my GPUs?

[Bubbleawsome]

Over the past 3 years I've had 3 GPUs die on me. (5 if you count returned cards)

The first was a OEM 5770 in a dell box. Died at about a year old.

Next was a 7770 on a 550w PSU. I OC'd it a bit too much, died about 6-8 months old.

The next two were 7970s on the same 550w PSU. I got the first one from a member of these forums, died about a month in. RMA'd, got the same card back, sightly improved, died in three days. RMA'd. Get a new/used (I can't tell) 280x. It's starting to die now I think. Getting "AMD driver has crashed and recovered" message.

Every single piece of hardware in this machine has been changed except the PSU, RAM, and HDD. Obviously only the PSU should be the only thing effecting this. I also undervolted my CPU and downclocked it a bit. It became unstable after about a week of those settings.

I can't tell if I just have extraordinarily bad luck or if something is wrong with this thing. I would suspect the PSU, but it got some of the best ratings of it's class. It is the rosewill HIVE 550w.

I don't have any money to replace any parts right now, but if you guys have any ideas on keeping the next RMA card alive I'll take them. Thanks!

 

[Rvenger]

Are you sure it is not your PSU? I mean, it can be faulty and its not out of the realm of possibility. I had a fault in my Seasonic Built NZXT Hale82 due to plugging the 24 pin into so many different motherboards, the pins started arching on the connector since they were so loose causing all kinds of erratic behavior.

I am running a HIVE 850w on 2x GTX 970s and this power supply has been great so far.

 

[SlickR12345]

550W PSU doesn't tell us much. If its a crappy Chinese PSU it can have 700W or more and still not be able to handle a powerful GPU.

I mean you can get a golden+ PSU that is 500W that is going to be able to run most GPU's, even something like the GTX 970 or whatver, but if its a Chinese 550W PSU then that is the problem right there.

 

[Piroko]

I had a 6670 die on me because the RAM chips and VRMs got scorching hot. My old case was terrible in funneling air to where it was needed, after some testing with different setups it was quite obvious that the lower half of my case was a dead spot in air circulation.

Also, for reasons unknown to me, that particular model had its heatsink expand over the RAM chips, but no thermal pads connecting them. That probably made it even worse.

 

[RussianSensation]

Do you have a surge protector? Maybe it's faulty/old. Sometimes sudden spikes in current/voltage spikes over time degrade PC hardware most sensitive to fluctuations. GPUs are very sensitive. If you don't have a household with stable electricity and run your PC without a surge protector, your GPUs are slowly dying and will keep dying. There are special tools which can measure how stable the current/voltage in your electrical socket are.

"
Excessive voltage will cause breakdown and arcing, either between separate exposed contacts (especially little points of solder), or through an insulating or dielectric barrier (for example, the dielectric in a capacitor).

Excessive current is generally caused by excessive voltage. Passive components have a specific, fixed resistance; the current through them is determined by Ohm's law, I = V/R. Excess current can cause overheating and burn-through of components, leading to open circuits.

Once you get into active components (piezoelectrics, op-amps, transistors), then the effects of going outside the components' specifications can be more complex, but never good."
http://www.instructables.com/answers/Does-Voltage-Or-Current-Kill-Electronics/

 

[Flapdrol1337]

A modern psu should always produce normal voltages even with input voltages way out of spec. Or are we talking ancient psu?

 

[Piotrsama]

Yes, sounds like the PSU is failing.
Time to buy a new (and better) one, and get one with a higher Wattage, as it's better when the PSU does not work at or close to 100% capacity.

 

[Bubbleawsome]

Do you have a surge protector? Maybe it's faulty/old. Sometimes sudden spikes in current/voltage spikes over time degrade PC hardware most sensitive to fluctuations. GPUs are very sensitive. If you don't have a household with stable electricity and run your PC without a surge protector, your GPUs are slowly dying and will keep dying. There are special tools which can measure how stable the current/voltage in your electrical socket are.

"
Excessive voltage will cause breakdown and arcing, either between separate exposed contacts (especially little points of solder), or through an insulating or dielectric barrier (for example, the dielectric in a capacitor).
-snip-

It is an old one. From the mid 90s I think. Might should get a new one.

550W PSU doesn't tell us much. If its a crappy Chinese PSU it can have 700W or more and still not be able to handle a powerful GPU.

I mean you can get a golden+ PSU that is 500W that is going to be able to run most GPU's, even something like the GTX 970 or whatver, but if its a Chinese 550W PSU then that is the problem right there.

It's a rosewill HIVE 550w. Got pretty good ratings. http://www.anandtech.com/show/5674/rosewill-hive-550w/6

I had a 6670 die on me because the RAM chips and VRMs got scorching hot. My old case was terrible in funneling air to where it was needed, after some testing with different setups it was quite obvious that the lower half of my case was a dead spot in air circulation.
Also, for reasons unknown to me, that particular model had its heatsink expand over the RAM chips, but no thermal pads connecting them. That probably made it even worse.

It's got great airflow. 120mm fan right to it. Even under F@H load it stays under 65c with the default fancurve or under 75 with fan set to 35%.

It is a gigabyte 280x with the WFx3 cooler.

I might have to keep RMA'ing cards until I can prove it is the PSU. I don't have money to replace it and will not for a while. I was planning for my next jump to be into the 800w range so I could CF another 280x if I wanted to.

 

[Piotrsama]

It is an old one. From the mid 90s I think. Might should get a new one.

Probably better without it, just connect your PC directly to the wall.
Most actual PSUs already have surge protection integrated, and according to the review yours also has it:

a varistor in the entrance is a good idea

It's a rosewill HIVE 550w. Got pretty good ratings. http://www.anandtech.com/show/5674/rosewill-hive-550w/6

It's not very amazing.

The power supply uses a well known layout from Sirtec, and it's a cheaper design. The PSU is nothing special, corresponding to what we have seen in many other models. The EMI filtering is well equipped, and a varistor in the entrance is a good idea. All cables should be protected with heatshrinks, and while some of them are we would like to see the others protected as well. The relatively low-end Teapo capacitors on the secondary side are acceptable for a PSU in this range, while the fan is pretty low-end as well. Rosewill promised to use a Japanese main capacitor but this time they used Teapo.

 

[RussianSensation]

Ok start testing your 12V and 3.3V rails under load then for 24-48 hours. Start HWMonitor running and do stress testing on the CPU for 12 hours. Record the lowest and highest 12 and 3.3V rails. Spikes or drops can be captured this way. Perform the same testing with the GPU loaded. Sometimes a crappy PSU gives out low 12V of 11 or 3.14V on 3.3V. If so, it's the problem.

 

[KaRLiToS]

Refer to this post to test the PSU different rails using a 100% accurate tool called the multimeter

http://www.overclock.net/t/1392568/new-build-questions-about-psu/0_30#post_19997799

You can also test with a SATA power connector

Tolerance table.

 

[Bubbleawsome]

Will do karlitos. Software-wise the only voltage with a weird spec is the +12v up at 13.5v. Otherwise they are all in-spec, though at the very top.

I hope I don't melt anything.

 

[KaRLiToS]

With aida64, my +3.3V rail indicates 2.9v but when I test it with my multimeter, it gives a 3.3v. That is the only reason why I would prefer to use a multimeter.

 

[liverunnin]

A 7970 crossfire on a 550w psu, is that even enough?

 

[Bubbleawsome]

A 7970 crossfire on a 550w psu, is that even enough?

No, it's just one 280x on a 550w. If I do replace this PSU I plan to go ~800w and CF later.

 

[Spungo]

Ok start testing your 12V and 3.3V rails under load then for 24-48 hours. Start HWMonitor running and do stress testing on the CPU for 12 hours. Record the lowest and highest 12 and 3.3V rails. Spikes or drops can be captured this way. Perform the same testing with the GPU loaded. Sometimes a crappy PSU gives out low 12V of 11 or 3.14V on 3.3V. If so, it's the problem.

Download OCCT. I don't know if the current version does it, but I know version 3.x would give very good data about voltage ripple. A bad PSU was very obvious because the voltage would be all over the place.

A really screwed up PSU looks like this:

A good PSU looks like this. Notice the voltage ripple %

 

[westom]

Software-wise the only voltage with a weird spec is the +12v up at 13.5v. Otherwise they are all in-spec,

Voltages can be in spec and still be reporting a defect. 'Where in spec' adds additional information. Post actual numbers to learn what those voltages are also saying.

For example, based upon what was posted, 3.2 volts on your 3.3 volt rail may be reporting a major defect.

Voltages must be measured with a meter and to three digits. All digits are significant. Motherboard measurements may be inaccurate until calibrated using a meter.

Also essential are voltage numbers from the PSU's green, gray, and purple wires.

 

[bononos]

Will do karlitos. Software-wise the only voltage with a weird spec is the +12v up at 13.5v. Otherwise they are all in-spec, though at the very top.

I hope I don't melt anything.

That would be 12.5% out of spec which is above the atx standard of 5%.

 

[RussianSensation]

13.5V on a 12V would kill any videocard with time. That's way above spec.

 

[cusideabelincoln]

The Hive has a 3 year warranty so maybe you'll luck out and they'll replace it, because it is most likely the culprit. Otherwise, just have some peace of mind and replace the PSU anyway.

All of these components have tolerances. Obviously going beyond those tolerances will surely lead to problems immediately or in the future, but even skirting on the edge of acceptable can lead to the same result.

 

[Charlie98]

13.5V on a 12V would kill any videocard with time. That's way above spec.

^^^^ Yaa, that's a pretty sizeable spike!


OP, I would yank that PSU... I know you don't have any money, but you can get a reasonable Corsair CX500 to tide you over for $30 (AR), and then keep it as a backup or sell it when you go bigger.

I also have to ask, based on this...

Next was a 7770 on a 550w PSU. I OC'd it a bit too much, died about 6-8 months old.

...if you are getting silly with the OC? It may not be one single thing... but a bad OC, a bad PSU, and maybe a bad surge protector all together add up to disaster.

Definition of insanity is doing the same thing over and over and expecting a different result....

 

[skipsneeky2]

OP, I would yank that PSU... I know you don't have any money, but you can get a reasonable Corsair CX500 to tide you over for $30 (AR), and then keep it as a backup or sell it when you go bigger.

The CX500 is a pile of junk for anything moderate,at best in my experience its good for cards up to a 760/670/7850 type deal.I had one for maybe a year and when i plugged in my then new 770 the psu was cooking itself and the card did not perform as it should.A completely stock clocked system with a i5 2500 and a 770.

Ended up having to get a 600w psu before my issues disappeared.

 

[westom]

A modern psu should always produce normal voltages even with input voltages way out of spec. Or are we talking ancient psu?

That was even true with the original IBM PC. AC voltages can change so much that incandescent bulbs brighten by 50% or dim to 40%. Those are also perfectly good AC voltages for any properly constructed computer.

More numbers. 5 volt digital semiconductor datasheets clearly state that voltage can increase by 40% or 50% without harm. This might cause timing changes resulting in a software crash. Then a user speculated hardware damage due to timing changes.

13.5 volts is only a 12% overvoltage. Most 12 volt semiconductors can even be at 15 volts or higher without damage.

Also bogus is a myth about higher wattage for better quality. Often a higher wattage without a corresponding price increase means essential functions are missing. Functions that, for example, mean a power supply cannot cause motherboard or disk drive damage.

Most will recommend on brand name due to insufficient electrical knowledge. Fewer who know this stuff could say more IF all voltages were provided. Including numbers from a PSU's green, gray, and purple wire. And only if measured by a volt meter; not by motherboard hardware.

ATX standard says 5%. Adding other facts say why a measured tolerance is different. Better answers provides unadulterated data - the numbers - so that others who know more about this stuff can post informed conclusions. An accurate answer immediately and without any more RMAs.

Or just shotgun - replace a PSU hoping that will fix it.

 

[Silverforce11]

I stopped using cheap low quality PSU a long time ago when one exploded, taking along my entire rig with it.

Now I still have a Silverstone 750W that's 6 years old and its still purring along silently even at max load.

Always get a good PSU first for any build.

 

[Bubbleawsome]

^^^^ Yaa, that's a pretty sizeable spike!
OP, I would yank that PSU... I know you don't have any money, but you can get a reasonable Corsair CX500 to tide you over for $30 (AR), and then keep it as a backup or sell it when you go bigger.
I also have to ask, based on this...
...if you are getting silly with the OC? It may not be one single thing... but a bad OC, a bad PSU, and maybe a bad surge protector all together add up to disaster.
Definition of insanity is doing the same thing over and over and expecting a different result....

Checking with occt yielded normal results, but I didn't have much time. It's finals week and I'm not doing great in my classes. D: I'll get the multimeter test by monday hopefully. The electricity in this room is horribad. The lights dim a bit if I have the mac and PC at full load and flicker if the ancient laser printer turns on.

The OC was a 1400Mhz OC. I got it up to 1500 for a bench and it died. Even stock 1000 was too much. Golden card too as it wasn't power throttling or going over 70c. ln2 might have made a record with it.

That card is gone now and I have a stock 280x and undervolted 4670k.

I think (hope) 12v is wrong.