The Titan was notorious for going bad due to DIES on their chip dying, not just the horrible board-level design choices they made. Arguably the lowest reliability design anyone ever produced, though it was the long-time king of efficiency for Scrypt when it DID work.
I disagree. Run Titans at 60mh per cube and they will run pretty much forever. The errors came when people ran the things at 80mh, which caused:
1) Dies to fail from too much heat. I have seen boards with burn marks on the back.
2) Power supplies running too hot. Once again have seen supplies with caps and FETs that caught fire.
3) FPGAs on the controller blowing up: This is caused partially by a crap design, part because people unplug the miners from the controller while running.
4) Removing and replacing the heat sink without new thermal compound (it's a one time use).
5) Running miners in barns, sheds, and places where either there is not enough cool air or there is literally horse crap everywhere (oh yes, I have had to clean those miners....)
6) Torquing the bolts so tightly the board tacos (this can be addressed by putting one of those rubberish thingies under the board between the front two posts).
7) PCIe power plugs burning due to crappy power supply cables or running too hot (don't use a little supply with cheap wires. Bad)
They weren't the best miners in the universe, but they weren't too bad. With the exception of 3 (and that's a biggie, using fucking optoisolators would have made everything ok) they're pretty good at 60mh, <42c on the die, under 70c on the supplies. Exceed that and things get more complex.
YMMV.
Per KNC's stated specs, they appear to have *supposedly* been designed to run at 80 Mhs - like the A2 was "base clock" at 1000, but if you upgraded the OEM power supply they have proven to be rock solid reliable at 1200 (the OEM power supply was only rated for 1100 watts, which is comfortable at 1000 Mhs clock but VERY marginal at 1100 and NOT ENOUGH at 1200).
The PCI-E connectors even at the 60 Mhs level were very marginal for the power draw, though not much worse than the Spondoolies SP20 (which was NOT notorious for flames as it had power limiting in the controller).
A LOT of the reported fails were on machines that the user had NOT done any form of "modifications" to.
A2s are quite comfortable running in high ambient temps in "turbo mode", they have some SERIOUS airflow through the cases and lots of heatsink for the design power dissipation (though the 2 fans in the rear of the case were a waste, nice to unplug them to have around as spares though).
Compared to the A2, the Titan was JUNK on reliability (except the A2 original power supply if you pushed it too hard), mostly due to BAD DESIGN DECISIONS that required folks to run them WAY too hard to meet their stated hashrate spec.
This also wasn't new for KNC - at least 2 of their SHA256 miners had the same sort of issues, the Neptune might have even been WORSE on the same points.
