That doesn't sound right.
a) You can put non-ECC memory in a Xeon board - Even the MP line. You just can't mix and match is all. On the other hand you usually can NOT put ECC modules in a non-ECC capable system.
b) Software doesn't care or even know about ECC. That's a hardware (embedded controller) convention. All memory reads are checked for bit errors when ECC modules are used (in an ECC capable system). It's only checked for single bit errors tho. If two bits are in error it gets passed into the system as being good - and will cause corruption.
Bit errors happen more or less frequently depending on where you are on the planet. Most of the errors occur due to cosmic rays which we're constantly being bombarded by. Most of those rays emanate from the center of our galaxy which is a thin pancake shape. And so the nearer the equator one is the more of these cosmic events will occur. Additionally radioactive materials and some forms of background radiation can cause bit-flips (errors) as well. Operating temperatures, physical silicon area, and die density all play a roll in the susceptibility of memory to bit-flips from the above named causes. Probably in the very worst cases a large system may experience as many as one to five such errors per week. In a typical desktop system with high density 16GB modules it's probably more like 1 to 3 detected/corrected bit-flips per year.
The only cons of EEC memory are the price difference and about a 1 to 2% performance hit (if that). You really DO want ECC if you can get it. Using non-ECC memory is like requesting that your system be less than 100% correct.
Last edited:
