Steve told an interesting story today. See, I've been running these enormous DMRG calculations on our local computing cluster. Sometimes the data comes out a bit funny. There's several reasons why this happens, but here's some of the more entertaining:
No joke. If a cosmic ray comes in and hits one of your memory units in the right way, you get a loss of data on that one bit. This happens especially for longer calculations, but there are ways of managing the errors in pricier computing chips.
Heat can also do this. Apparently back in the early days of computing if you suspected your results were wrong because of this error, you ran a diagnostic, took a panel off the side of the computer, and held a blow dryer up to scan across each memory unit. If you got an error, with your blow dryer over this spot, you knew the bad node was in there.
Yay modern computing! I don't have to do that!