TheScullster wrote:
> Hi all
>
> Recently bought 3 x DT790 Dell desktops, one of which suffered "Windows Shut
> Down Unexpectedly and Needs to Recover" type errors and also, Kernel Power
> event id 41 issues. After swapping components around with the other 790s,
> the problem was seen to follow the RAM. So Dell are sending out
> replacements. Dell's supplied diagnostic showed no hardware issues at all.
>
> OK so this failure in itself is not so surprising. The part of this
> situation that I find strange, is that the erratic behaviour occurs almost
> exclusively overnight. Between 21:45 yesterday and 08:18 today, there were
> 44 errors reported, with the last one (at least) resulting in a machine
> restart which I witnessed as I arrived at my desk. The longest time period
> between errors was approx 1 hour and the shortest just a few minutes. The
> PC has been up for three hours now with no errors reported.
>
> Now I know from server UPS data, that our building is subject to
> over-voltage on supply during the night. Is this likely to give rise to
> this erratic behaviour? When testing the problem PC, the other 790s were
> also left on over night and these reported no errors (the desktops do not
> have UPS on supply).
>
> Comments appreciated.
>
>
> Phil
You'd look at the UPS make and model number, and look up the specs.
Chances are, it conditions the line, and adjusts the output voltage
such that the ATX supply doesn't see anything. (There are at least
five different architectures of UPS design, so YMMV.)
And the ATX supply has its own resiliency. An ATX supply can operate
down to around 90V. And the ATX supply has hold-up time, provided
by the main capacitor inside the supply. When a PC is idling, this
hold-up time can be significant (up to around a second or so). You'd
practically have to kill incoming AC to the UPS, and wait for the
UPS battery to drain, before that desktop PC would notice.
That's a guess.
*******
http://support.microsoft.com/kb/2028504
"Description of Windows Kernel event ID 41 error in Windows 7
The system has rebooted without cleanly shutting down first"
Once you get the new RAM, try running Prime95 "torture test" for
four hours, and see if no errors are detected. You should really
do "acceptance testing" on new PCs, because the factory testing
can amount to a 2 minute run. Dell would rely on the RAM manufacturer
to test the RAM, before it's assembled. And we all know how
the RAM manufacturer makes money - by reducing test time. That
means the "consumer is the tester". Prime95 is a good acceptance
test, and will give some idea whether the PC is ready to deploy.
http://www.mersenne.org/freesoft/ ftp://mersenne.org/gimps/p95v266.zip
Extract file (no install needed). Keep the files in a separate
directory. Double click "prime95.exe". In "Welcome to GIMPS"
dialog box, click "Just Scress Testing". In "Run a Torture Test"
dialog box, "Blend" is good for a start. If you used the 64 bit
version, it's possible using "Custom", you could select a larger
memory test size. On the 32 bit version, you may be limited to
around 1600MB. In any case, start with "Blend" and open Task Manager
after it starts running, to see what resources it is using.
When the four hours is up, if all icons are still "green", select
"Stop" and then select "Exit" from the menu. The program should
then be gone.
It is even possible to run multiple copies of Prime95. Create a
separate folder for each copy. Copy the prime95.exe file into
each separate folder. Start the program as above, and select
custom and set the number of test threads and memory test size.
By doing it that way, you could test more of system memory,
by using say four copies testing 1600MB each. You'd run fewer
test threads on each copy, to spread the load.
Note that the stress from this test, is too much for older OSes.
Before the computer runs out of resources, instances of the program
may die. Instances may die even before the system runs out of RAM.
I was quite surprised by that. It's only an issue, from a test case
design point of view (i.e. wanting reproducible behavior for the
test). I've only got one copy of Windows 7 here, and wouldn't
expect the same problem with that.
Still, Prime95 is a great shakeout test for a new PC, and you'll
get a test of its cooling system, as well as RAM and CPU. The
testing icons in Prime95 should stay "green". If a test thread
stops, you could have bad RAM or a bad CPU. Or in the case
of an enthusiast retail motherboard, something needs to be
adjusted for stability (like bumping Vdimm perhaps, or adjusting
tCAS).
Paul
Strange RAM Behaviour 
Linear Mode
