graphics card keeps crashing during bitcoin mining/opening folders,etc

[tarui]

i;m using a 6990 on 11.6 drivers running at 840/1250

power supply is a seasonic x750

previously i was running at 880/1250 (bios switch 2) for a few days and one day it just keeps crashing within mins of mining, i reckon it's due to the psu because 1 review site says that a 6990 running at 880/1250 at load draws 660w. plus the temps were frequently above 90 or 95C

so i ran it at 830/1250 and it was fine. so manually overclocked it back to 840/1250 and today it keeps crashing for no reason again. i kept the temps below 90 or 85 if possible as much as i can by using fans to direct cooler air to the pc.

the gpu-z's logs shows the following results.
800mhz changes to 0.4mhz
mem clock 1250 changes to 3200mhz
temps 81 changes to -1 C
fan speed 30 changes to 100%
fan rpm 3700 changes to 0
temp 2,3,4 80 changes to 32767C
vddc changes from 1.12v to -
vrm temp 78C changes to -


just a moment ago, while it was mining along fine, all i did was open a folder on the desktop and the system crashed again.

when i tried to downclock back from 840 to 830 using sapphire trixx, it says

amd ultra low power state is enabled for this cf configuration. overclocking a ULPS enabled card will crash this system.
 go to settings, disable ulps and restart system to enable OC.

i downclocking, not overclocking!

[Trillium]

Wow I typed up a 2-page response to your post and then the stupid forum auto-logged me out and I lost everything. I will give a brief re-summary.

- I have a lot of experience with 5970's, the dual gpu card of the 5xxx series
- You can disable ULPS 'completely' by ticking the check box in trixx but then also running a batch file / program called ULPS_Configuration_Utility_v1.1.4 that you can download. You need to restart after toggling the state. When you are not overclocking and the card is stable you shouldn't have to go through the ULPS nightmare again.
- Booting into windows and getting folders crashing means you have either permanently damaged the card or you are running a unrealistic BIOS or you have a program forcing unrealistic clock speeds/voltages as soon as windows boots.
- When GPU-Z displays values like you copy-pasted on a dual gpu card it means that one of the cards has hard-locked and you must restart the computer to reset it.
- You should instead run the card at sub-stock speeds and test with GPU testing in a program called OCCT. Make sure you have error checking on, the card is installed as the primary display device, and you are running the test with highest shader setting and fullscreen (to get 2x100% utilization of the cores). Run it with increasing speeds/voltages until it crashes or gives errors. Never change speeds/voltages while the tests are running.
- TO test the PSU, run the OCCT CPU (>C<PU) test and keep an eye on the voltages it records. Then start your miner or any program that uses 2x100% load on the GPU cores(not a game...), like fullscreen FURMARK or something other than OCCT. Stop that program and then also end the CPU stress test and peruse the voltage plots it has saved. As to how much the voltages should vary when you amped up the graphics card then I would say never more than 2-3% in my opinion. For my system I can pull 700 watts in under 1 second from my Corsair 1000 watt PSU and the 12v rail voltage and all the others will not vary by more than 0.01 v, it all depends on the size and quality of the output rail capacitors in your powersupply.

[legitnick]

i;m using a 6990 on 11.6 drivers running at 840/1250

power supply is a seasonic x750

previously i was running at 880/1250 (bios switch 2) for a few days and one day it just keeps crashing within mins of mining, i reckon it's due to the psu because 1 review site says that a 6990 running at 880/1250 at load draws 660w. plus the temps were frequently above 90 or 95C

so i ran it at 830/1250 and it was fine. so manually overclocked it back to 840/1250 and today it keeps crashing for no reason again. i kept the temps below 90 or 85 if possible as much as i can by using fans to direct cooler air to the pc.

the gpu-z's logs shows the following results.
800mhz changes to 0.4mhz
mem clock 1250 changes to 3200mhz
temps 81 changes to -1 C
fan speed 30 changes to 100%
fan rpm 3700 changes to 0
temp 2,3,4 80 changes to 32767C
vddc changes from 1.12v to -
vrm temp 78C changes to -


just a moment ago, while it was mining along fine, all i did was open a folder on the desktop and the system crashed again.

when i tried to downclock back from 840 to 830 using sapphire trixx, it says

amd ultra low power state is enabled for this cf configuration. overclocking a ULPS enabled card will crash this system.
 go to settings, disable ulps and restart system to enable OC.

i downclocking, not overclocking!

I had the Exact same problem a year ago with my nvidia card. I ended up replacing my graphics card and the problem was fixed.

[Trillium]

Yeah it's not too hard to work out if a card is dead. If its under warranty then a RMA is the way to go. If it's out of warranty you could sell it on ebay for parts (surprisingly good money doing that, since some people like to claim they can fix them). Of course you could try baking the card if you suspect that the problem might be a solder/connectivity problem on the board, but that is essentially a 'last resort'.

[tarui]

it gets worst.

crashes the moment i start furmark.

[Trillium]

it gets worst.

crashes the moment i start furmark.

I would try the card in another computer (friends computer? spare computer?) just make sure it has a suitable powersupply and you remove the old card's drivers first.

Make sure your using original/stable bios and not overclocking past manufacturers stock speeds for the test.

If it still fails the tests then I would suggest doing an RMA or if it is no longer under warranty try underclocking it until it is stable and/or remove the heatsink fan assembly and check for any signs of damage or poor thermal contact (dried or cracked thermal transfer material over cores, or more likely the VRM MOSFETS.) Also you can get a magnifying glass and carefully inspect all the tiny SMD capacitors and resistors that are all over both sides of the cards for damage (from mechanically breaking them off) or in the odd case of them exploding (which is a problem on some cards, eg. overclocked 67xx and 57xx cards in particular). If you find a damaged component it is not impossible to repair it yourself. If you do not have the skills/knowledge to do these kinds of things then I'd suggest pawning the card off as a 'faulty / parts' auction on eBay, you should get at least a few $$$ for it.

Edit: Also inspect the PCI-e (motherboard) connector pins for damage. If they appear dirty, clean them (google). Check the PCI-e power connectors and PSU cable plugs for signs of burning, if some of the pins/cables have charred then current will most likely not pass through it sufficiently and the card will be unstable, this kind of thing can happen - not surprisingly - on high-power cards like 5970, 6990, 7990 etc.