I had a similar problem on a non-PC computer - turned out to be a weak PAL
chip in the memory addressing circuitry.  At first I thought it was bad cache
memory, or the parity generator, but I still kept getting random parity
errors and bad sector errors on the Hard Drives.  The problem was that
sometimes a "high" wasn't "high enough" and was misread as a low.  When
the SCANDISK equivalent I used tried to check the data at that address
it got the wrong data and assumed that that sector was bad.

> From:    Mike Wood <[log in to unmask]>
> A friend has an old CTX with a 100Mhz pentium overdrive cpu, 48 megs of ram,

I gather that everything is "out of warranty", and not a disaster if you
"drop the ball" while trying to locate the problem...

> award bios and win95. He tried so set up an internet connection but was missing
> to many files and drivers, I advised him to upgrade to win98se which he tried to
> do. He made it to the point where he had less than 10 min to go and it started
> rebooting with error messages and finally locked up completely. It would power
> down and run scandisk and then hang.
> I brought it home, did a deltree windows, and did a complete install of win95b.
> this worked fine for one day, and then started running scandisk and repairing
> files on evety boot. We decided that the only alternative was a format and
> reinstall. He used a boot disk that had been created with win98. the format
> kept reporting on the number of allocation units reclaime
> d (?) and after 24

I assume that this means he used the CORRECT low-level format from the drive
manufacturer - and it found that after exhaustive testing, each sector that
had been marked as "bad" by SCANDISK was actually OK.  This would be
expected to take much longer than the usual check of a "good" sector, as
it has to test for all possible problems to prove that there is nothing
wrong.  If something was physically wrong with a sector, the first test
would probably spot it.  This doesn't guarantee that there is no physical
problem with the drive - a bad bearing, etc., could cause random errors, too.

I have read of low-level formats that took days to complete...

A thought - seeing as the drive is now an expensive paperweight - you might
try low-level formatting the first 10 cylinders (out of 1023 or whatever).
leave everything else (such as parking cylinder) the same as before.
This should quickly prove that it works by giving you a small (say 10 Meg)
drive that you can FDISK and FORMAT/s.  You will, of course, have to do a
full low-level format to use the whole drive. This may be quicker than
the last time, if most of the disk has already been checked.  Does your
friend remember the cluster/cylinder number when he aborted it?  It may
also go faster if you use a known good PC.

> hours was still formatting, 1gig hd, he paniced and cancelled the format and
> tried again with a win95 bootdisk. Now the format command comes back with
> invalid disk.
> Can this hd be salvaged and if so how? Any help will be greatly appreciated.

I would suggest using another PC to try to recover the drive (no hope for any
data - it was randomly corrupted before the aborted format).  Try a small,
old (but known good) hard drive on the problem PC.  Odds are, it will
*seem* to go bad in the same way (and should be fixed in the same way.)

Wherever you can, try to divide components that work/don't work reliably.

Use known good SIMMs, cache SRAM, etc. in the same way to try to locate the
problem - just don't mess up which part is "good" and which is "suspect".

If you have an oscilloscope, try to find "weak" (lower-than-usual peaks)
on the address lines at the cache/SIMMs/HD cable.  Often, just the extra
load of the probe (typically 15 pF, 10 MegOhms) will pull the line with a
"bad bit" low.  The system may then seem to work by consistently reading
the same (wrong) address. Use a small point on the probe, insulated
except at the very tip, so that you will avoid short-circuits even if
the probe slips between pins on a chip.

Good luck - these intermittent faults are real "horror stories".

Boyd Ramsay

[log in to unmask]

         PCBUILD maintains hundreds of useful files for download
                     visit our download web page at:
                     http://nospin.com/pc/files.html