[ILUG] [Slighty OT] ECC memory in desktops
felix at compsoc.nuigalway.ie
Mon Jun 14 13:55:40 IST 2010
On Fri, Jun 11, 2010 at 08:07:36PM +0200, Oliver Pfaff wrote:
> As far as I know ECC is currently reserved for server applications and
> almost all desktops/laptops use non ECC.
> If the failure rate was that high or ECC helped so much you wouldn't be
> able to buy a non ECC system.... I'm open to correction on all these
> I also believe that the selection of ECC is limited and that your limited
> in the top speed of the EEC memory modules currently available. I don't
> think komplett offer ECC memory at all.
True, DDR3 PC-10666 is the fastest currently available, but then that's
the max speed unless you plan on overclocking, which I don't.
> Also your mainboard needs to support ECC so you'll pay extra for the
> mainboard, extra for the memory and take a performance hit.
There's a few motherboards for AMD processers in the €100-150 price
range, so I don't think that will be too much of a problem. The extra
price of the top speed DDR3 ECC ram will make a noticable difference.
> >From the article:
> "I computed that if you have 4 GiB of memory, you have 96% chance of
> getting a bit flip in three days because of cosmic rays." Is he assuming
> that all 4 GB are in use? Right now my Linux Mint system(Gnome) is using
I did mention that there were problems with the first article.
> > First blog entry is contains some inaccurate statisical analysis on the
> > error rates, second one seems closer to the mark.
Any yes, utilization of memory would be a factor. Definitely the DRAM
study indicates that error rates, when they occur, and memory
utilization are connected.
> I would argue that there are about a billion personal computer sin the
> world and if non ECC was causing such a problem we'd have heard about it by
Do we even have reasonable statistics to compare? How often have kernel
panics (BSODs) and various intermittant software crashes on systems been
written off as software bugs that remain unreproducable on any other
hardware? Not to mention that if the systems are not heavily utilised,
they have a much lower chance of hitting issues.
How often have people seen corrupted files?
How often do Window's users reinstall systems, or recover a broken
system using system restore and reinstall all the same software and not
encounter a similar problem for months more on end?
I'm less inclined to believe that even with a billon desktops not using
ECC that the problems associated with memory errors would actually be
attributed correctly to faulty memory.
"Nothing is foolproof to a sufficiently talented fool."
More information about the ILUG