After the fiasco of nVidia and 196.75 the design of graphics cards has changed significantly. Older cards like our Zotac 8600 GT and our EVGA GTX 260 SC have thermal sensors but they lack functionality to downclock the device to prevent self destruction. More recent cards however are able to throttle back to save themself from damage.
The EVGA GTX 260 began to run very hot and it was close to the maximum 95°C. We removed the blower assembly and applied new Arctic MX-4 and reassembled it. The card then ran dramatically cooler suggesting that the OEM thermal pads had degraded.
EVGA has told us that repairing the thermal material does not violate the warranty on their cards. EVGA considers it to be field repairable. A large 20g tube of Arctic MX-4 is enough for over 12 video cards.
nVidia Fermi cards like our Asus GT 630 do not have a thermal throttle so its important to monitor temperatures. MSI Afterburner can make a custom fan profile to be more aggressive if heat buildup is an issue.
nVidia Kepler was the first series to have practical thermal throttling to prevent the GPU from destruction. Cards like our EVGA GTX 660 Ti have the new dual fan cooling solution and it has never overheated on us even with the heavy gaming use.
AMD GCN architecture have thermal sensors to protect the card. Our Asus HD 7870 also has a dual fan cooling assembly and the card has also never been problematic.
A good example is our Gigabyte GTX 750 which began to throttle under extreme load. When we removed the GTX 750 cooler we cleaned it up and applied new Arctic MX-4 and reassembled the card. The card became thermally stable and it no longer was throttling back. This is a good example of a card that can survive until it is repaired.
The basic skill of repairing thermal material can add many years to the service life of the hardware. The GTX 750 only 4 screws while larger cards like the GTX 260 have 8 screws. They are easily identified with the springs on them.