Comment 119 for bug 346691

To build a better understanding of the mechanism, it's worthwhile finding out:
 - is there sufficient cooling for the southbridge and northbridge?
 - are you running the latest BIOS?
 - are you running the vendor's validated BIOS defaults?
 - is the powersupply of reasonable quality/spec
  -> for lower ripple and supply rails within tolerances
 - the output from 'lspci'
 - importantly: can the corruption be provoked in MS Windows?

I've experiences two cases where bad memory has been exposed though fast I/O (in a HPC environment), but memtest didn't detect issues, so there is still a small chance.