[conspire] cd burning giving slight errors

[Rick Moen]

Quoting Peter Knaggs (peter.knaggs at gmail.com):

> I eventually tracked the problem down to a bad memory slot
> on the motherboard itself.  Apparently that one slot isn't capable
> of running at full speed, no matter what memory stick I put
> into it.  Short of getting the manufacturer (ASUS)
> to replace the motherboard (It's still under warranty, it's
> a P7P55LX), I've found that running the entire
> memory bus at a slightly lower frequency setting
> (800MHz instead of 1333MHz) solved the trouble
> of reading and writing CDs. It's curious that memtest86
> doesn't show any troubles, and parallel linux kernel builds
> went through fine: it was only I/O to the CD drives that
> was affected by this "bad memory slot" issue, the symptom
> being I/O errors when higher speed reading or writing was used.

That's pretty subtle, all right.

I'm going to make a silly wild-assed guess that the difference between
the memory activity exploited by memtest86 and the memory activity
exploited by CD-reading somehow involves motherboard circuitry used for
DMA in the latter case but not the former.  

As a reminder, I just don't have as much faith in memtest86 or any of
its immediate kin as I do in massively parallel (make -j) iterative
kernel compiles, as discussed in the conspire thread linked from my
http://linuxmafia.com/~rick/ personal page (concerning 'the best way to
test RAM', or words to that effect).

I remember a really subtle motherboard problem back in late 386 days.
I bought an offbrand motherboard with an AMD 386/40, which was great
with OSes du jour until I upgraded it from (if memory serves) OS/2 3.0
Warp instead of OS/2 2.1, whereupon it suddenly had major stability
issues that went away if I either disabled CPU caching (ick!) or stepped
down the system clock from 40 to 16 MHz (ick!), whereupon it was fine.

Eventually I figured out the cause for this wacky problem:  The OEM had
taken a motherboard that was in-spec for an Intel 386/33 chip and just
crammed an AMD 386/40 into it for higher speed appeal and slightly more
retail margin.  (They jobbed off the removed Intel chip separately, or
it didn't have one.)  Result was a motherboard that was just slightly
out of spec on the northbridge, leading to subtle timing problems with
some OSes but not others.

I junked the motherboard and made tried to avoid similar errors
involving hidden hardware weaknesses.