Hi,
some background info:
The databus, from the SDRAM to the processor is 64 bit width (normally 8 memory chips on one DIMM, and every chip has an 8bit width memroy bus 8x8 = 64). There is a possiblity to detect errors with a parity function. This means, that you have an additional Bit per 8 databits (==> you have 9 chips on your DIMM (or 2x9=18 for a two bank DIMM). with this additional bit the chipset can decide whether the data read from the DIMM is correct. there are several algorithms, that allow you to detectcorrect different kinds of errors (often correct single bit, and detect doulbe bit errors). Of course, the possibility to get an error is 1/8 higher with 9 devices than with 8 (the parity device can include an error too).
SocrPlyr's suggestions are the right way,
There is a (VERY) dirty solution too: disable the ECC in the BIOS when possible. If the failure is happening in the parity device your problems are gone (not very likely, and if this chip is on its edge, the others are close to a failure too).
If the error is happening in one of the data storage chips, the system is not able any more to detect the error, and is working on with the wrong data. Than it depends what is stored at the failing cell. If its part of the picture then you will have set one bit wrong in your picture, If it is part of the system/program code your system behaviour can be very inpredictable.
recomendation: reduce the FSB a little bit.
ruckb