That's indeed interesting and something I've thought about a lot. Here's a data point. I run a very small, 3-card (all XFX Vega 64) mining rig, which I was lu

That's indeed interesting and something I've thought about a lot. Here's a data point. I run a very small, 3-card (all XFX Vega 64) mining rig, which I was lucky to have started buying around early November before the price of graphics cards really started to explode. As a miner, you want to use as little power and generate as little heat as possible. This means in my case that my cards are run underclocked and undervolted. Two cards in my rig run at 1250MHz core clock and 1085MHz memory clock, at a voltage setting of 0.810V. I haven't been quite as lucky with the remaining third card in my rig - I can only run that at a core clock of 1220MHz and a memory clock of 950MHz, at a voltage setting of 0.850V. I have what some might consider rather aggressively low temperature targets for the cards - 54 degrees C. My two good cards can achieve this temperature with fan speeds around 3650 RPM. The fan on the not-so-lucky card needs to spin around 4500-4650 RPM to achieve this. This is on a hot day and these fan speeds are lower if I'm running my air-conditioner. I haven't seen what the rig is like in winter yet. These numbers are likely the worst-case, since it's summer right now. In my mind, these cards are subjected to much less stress than cards that have been used for gaming, which will run hotter, at higher voltages, with much more thermal cycling, often overclocked. The lifetime of the fans is probably my only concern when it comes to a card that's been used for mining and is what I'd watch out for if buying used. Otherwise, I feel far more confident buying an ex-mining card than an ex-gaming card. Mining done properly seems like it should be far less stressful on a card than gaming, to me. It would be interesting to see if any kind of concrete data has been collected on this actually - anyone know if any of the big mining farms have published any reliability data (akin to those data centre hard drive reliability reports that get released occasionally)?