The special Error Correcting Code RAM in the Mac Pro store 64 bits of data plus an additional 8 bits of "syndrome" data. This is essentially permutations of Parity on combinations of bits in the word. The syndrome bits are devised in a way that means if a single-bit error occurs when a word is Read there is enough information to correct the error on the fly (essentially no slowdown), and that is exactly what the Hardware does. This Hardware is fairly complicated, and is working VERY Fast. It is built into the high-powered Xeon Processors used in the Mac Pro line.
An error also sets a bit that indicates a correction has occurred, and where, and these are collected by a background process from time-to-time, and tabulated in that report when requested.
It also has enough information to determine whether an uncorrectable error has occurred. (Most double-bit errors are uncorrectable.) These will halt the machine with a very distinctive kernel panic, machine check, typically detected by multiple processors at once. This is by design, to keep the error from poisoning your data.
When single-bit errors are occurring, you are losing the protection normally provided by error-correction. Failure to deal with the issue and replace the failing modules may mean you will suffer an uncorrectable error, and the machine will halt. This more than a little bit inconvenient.
The error-correction logic is used in a special way during the Power On Self Test. In those first few seconds, ANY Errors that occur (correctable or not) will cause the module slots to be declared "Empty". The System will not use them.
