This is rather complicated but I'll try to explain as best I can. I had a Radeon 4850 fitted and 3 days ago I opened my PC to fit an extra Sata HD that I had laying around, 20 mins after I fit the HD the pc shut itself down. I powered on again and it shut down in 20 secs.
I waited 30 min, powered on and instantly loaded HWMonitor to check temps, my gpu temp was idling at 92c then 93,94,95.. it was going up. I shut down and opened the case again. Took out the card and made sure the fan was dust free and the card fitted ok.
I put it back in and watched carefully when I powered on, the fan WAS turning but only about 20% speed. I went into CCC and tried to adjust the manual fan control, I set it to 50% and hit apply.. no change, set it to 100%, no change.. set it to 0% no change!!? (have tried ati tray tools and MSI Afterburner too, both show the fan as whatever manual setting I put into it but in reality the fan neither speeds up nor slows down)
I thought perhaps the fan controller in the card was broken so I took out the 4850 and fitted my older 4670 card. it was fine and since then (touch wood) the pc has been fine.
Still, when I go into CCC and try to set the manual fan speed, although the computer says the speed has changed, it certainly hasn't changed. This fan is loud and no matter what I set it on, 0% or 100% it stays the same, idling speed.
I wondered if it was a software problem and today I uninstalled my graphics card drivers and downloaded and installed the latest ones from AMD.
No difference, I can set the manual fan speed but the fan itself has ZERO change. also I downloaded MSI Afterburner and tried to change the fan speed in that, and the software said the fan speed was changing but in reality it wasn't changing from idling speed.
I know for a fact that manual fan control did work as last summer when it was hot I manually turned my gpu fan up to 75% then when it got cooler I put it back on auto.
I did get a virus alert the day before I did this, could the virus have damaged the fan control or is there another reason?
Right now I'm stuck with this old 4670 and unable to play any newer games, I don't see the point of buying a new gpu if what is causing this is not fixed.
The first thing I would check is your device manager - check for any drivers that may not be installed properly (display drivers, mother board drivers, and anything with a yellow exclamation point on it).
Make sure you have the most current drives for all your hardware.
Also, check your BIOS settings to make sure that something isn't changed...
Mostly overclocking settings could be at issue. Also, the motherboard drives could be creating issues with the drivers for the GPU / GPU FAN. It is possible either a driver became corrupted or the BIOS flaked out on you.
You can test with programs like Prime95 (for stress testing computers). 99% of troubleshooting any device is narrowing down an issue to one part (hopefully), and swapping out a single part.
When you don't have parts to swap out (or they are too expensive to just try them), often it is a good idea to "breadboard" the computer. This involves removing all components from the case, and putting together the most simple circuit you can (i.e. minimal RAM, mobo, power supply - and if you don't have on-board video, the video card) and power it up to see what it does. By removing the extra components, you are narrowing down what could be the problem.
If all works, you add more components into the "breadboard" until you have a failure. In some situations, you will find that a loose screw, or alignment of the mobo could be slightly off, creating a shorting of one or more components.
Also - in the breadboard setup, it is a lot easier to try different parts, as opposed to fishing everything through a case to replace one at a time.
BTW - I haven't found software that when loaded will say, "You _______ is bad." They give errors like "memory error" or "video error" that would indicate a problem with memory, but it still could be multiple components. I would take what they do as the whole system either passes or fails.
On a known good system, if you add a component and a failure is generated, then more than likely you have a single point of failure.