On 3 September 2008, Ted Unangst <[EMAIL PROTECTED]> wrote:
> On 9/3/08, Chris Smith <[EMAIL PROTECTED]> wrote:
> > On Wednesday 03 September 2008 09:04:01 am Dave Wilson wrote:
> > > If you find that the build test fails, and then find that memtest
> > > succeeds, then you can deduce that the problem lies with your
> > > hard drive
> >
> >
> > Only if memtest is infallible. I may be mistaken but I've long held
> > the opinion that while a memtest failure almost certainly means
> > defective memory, a memtest pass does not carry quite the same
> > weight.
>
> I had a computer with bad ECC that would pass memtest. It did make
> little notes in the BIOS log, but memtest itself issed no complaints.
> Attempting to compile something though would cause all sorts of
> mystery errors. memtest is hardly representative of real world usage
> patterns.
Yes. FWIW, according to a friend who is a hardware designer and
cuts open memory chips for a living, you simply can't test memories in
software. That is, you can prove them broken, but you can't reliably
prove them fine. You need some really expensive hardware for that.
Also, just like disks, there is no such thing as a perfect error-free
memory. So the answer to any conceivable test will be a statistic,
not a definitive true / false. The difference between memtest and a
hardware tester is how accurate this statistic really is...
Regards,
Liviu Daia
--
Dr. Liviu Daia http://www.imar.ro/~daia
