Mostly people are doing something like flashing-in a custom bios/overclocking/overvolting/running too hot, etc., ad infinitum--and cards go "pop!" at some point. But to be honest about it, I've actually never lost a single 3d GPU from any manufacturer that simply expired (ATi, AMD, nVidia, 3dfx, STB, and others.) Never lost a single card. Came close, however, with a 3dfx V5 5.5k circa ~2000-2001 or thereabouts...;) I had changed heat sinks and was overclocking the crap out of the card--when suddenly it just went blank--wouldn't put out a signal of any kind (VGA or otherwise.) Put the original heat sink back on, and had rebooted about ~6 times consecutively and was preparing to give up when suddenly--ta-da--I got signal! System booted, and never had a moment's trouble afterwards. That was the closest call I've had since I bought my 3dfx V1 circa 1997-8?. But I probably don't use a GPU more than 18 months-2 years max, if that long. So, I suppose just plain longevity could be a culprit.
Don't see how just a set of software drivers could burn out a card--even if the drivers grossly overclocked the card somehow--it would just lock up as opposed to dying, and would unlock as soon as clock/temps got sorted.
I do know, however, that as GPU manufacturing processes get smaller the chance for electromigration damage from overvolting and overclocking rapidly increases--but the same thing happens with cpus, too. When you get to 14nm best observe stock voltages and cooling much more stringently, imo, if you want the processor to last.
