On 21 Jul 98 at 13:50, STT wrote:

> Now I think the memory upgrade is the most likely offender :) But I
> am baffled by memory and system tests which report no memory error:
> I also tried DOS 6.22 HIMEM.SYS which reported no problem.

  I've seen reports before that Win95 sometimes has problems with
memory that DOS doesn't.  I have a theory about this:

  Dynamic RAM -- including FPM, EDO and SDRAM -- is relatively cheap,
but doesn't hold its contents for very long.  There's a background
process using one of the DMA channels, which visits every location in
RAM periodically, to "refersh" the contents (restarting the decay
process in each bit).
  There used to be a few "utilities" out there which claimed to speed
up your machine, which they actually did by slowing down this
background refresh process.  [On many machines, there's a safety
margin between the frequency of the refresh and the actual decay rate
of the RAM.]
  My hunch is that Microsoft has done a similar "speed up" in the
Win95 kernel.  I have no proof of this; I can only say that (a) it
could explain why Win95 sometimes has problems with memory that works
fine for all other tests, and (b) it's similar to tricks Microsoft
has played before.
  [I'm not certain whether the size of the installed RAM affects the
timing of the refresh -- it seems reasonable that it might.  It's
possible that IF Win95 does something like this, it determines at
install time how far it can be pushed.  So a fresh install with the
32MB DIMMs in place MIGHT fix the problem -- it's a procedure that
has fixed a lot of strange problems over the last three years.]

  Assuming that this is a reasonably good analysis of the problem,
you should be able to return your DIMMs as "does not work with Win95"
(i.e., the decay rate is a little too fast).  You may have better
luck by going to a higher grade of SDRAM, which will probably cost
you a bit extra.  [My local store, for instance, sells RAM in three
grades, by warranty:  90-day, 1 year, or lifetime.]

David G