undefined | Better HN

0 pointstomxor5y ago0 comments

I like when people back up their claims with numbers, but would you mind describing roughly what that 96% probability of error is based upon?

I understand altitude has some kind of proportionality to cosmic ray exposure, and number of bits will multiply the probability of an error.. I'm presuming there is also an inherent error rate to DRAM separate from environment. But what are those numbers.

0 comments

bitcharmer5y ago

Apologies, you're totally right. I should have linked to the source:

http://lambda-diode.com/opinion/ecc-memory#:~:text=A%20syste....

tomxorOP5y ago

Great thanks!

[edit]

Looks like the calculation was revised [0] after criticism:

> Under these assumptions, you'll have to wait about 33 to 600 days to get a 96% chance of getting a bit error.

What's more worrying is the variance, the above calculation is based on expected well behaved DRAM.. yet some computers just seem to have manufacturing defects that make the incidence of errors high enough to be a regular problem.

[0] http://lambda-diode.com/opinion/ecc-memory-2

j / k navigate · click thread line to collapse