How to diagnose occasional sudden resets?

Question

I have a Windows XP system, and have recently upgraded by adding 2 1GB sticks of RAM to the 2x0.5GB already present. Since then, about once per day (the system is used 8+ hours per day), the system has suddenly and unexpectedly reset. On a couple of occasions, the system has frozen completely, only responding to the power button being held in for several seconds to force power off.

Nothing at all ever appears in the system event log that might indicate a possible cause - everything seems to suggest business as usual.

Sounds like faulty memory - but memtest86+ says otherwise. A full test, taking over an hour, found no issues.

The next likely suspicion, then, is that I've knocked something while installing the RAM. Trouble is, everything I can think of to test seems fine. I've opened up the case and prodded a few things around, hoping to get better contact on connections etc, but there's no sign yet as to whether that has made a difference or not.

I thought about a malware-related timing fluke, but again, so far as I can tell I'm all clear.

All I can think of to add to my checklist (mainly of things that I can't easily check) is...

• The power supply is (1) only 350W, (2) not necessarily the best quality, and (3) powering a Prescott P4 640 3.2GHz. Could that be borderline overloaded or about to die? How do I check?

• Is it possible that the CPU isn't getting cooled properly? I haven't had the fan past normal tickover even doing video encoding, and the only sane temperature reading from SpeedFan is pretty steady at 36 celcius, so probably not.

Any thoughts? Is there a standard procedure for diagnosing this kind of fault?

EDIT

I did finally resolve this issue, and the PC is now working reliably (has been for weeks).

The problem was the heatsink - a combination of two issues, really. It's a socket LGA775 processor, meaning the huge heatsink which locks into the motherboard via four "legs". One of the locks wasn't locked properly, probably meaning I'd knocked it when installing the RAM. This means the heatsink probably wasn't making good contact with the processor over its whole surface - ie one corner of the processor could easily have been running hot.

Second issue - the thermal paste had dried and crumbled, so it was basically thermally insulating dust.

I now have a complete new cooler - that wasn't really necessary, but the new one runs quieter.

Answer 1

From that description I would suspect the PSU is either on the edge of its specification or is starting to fail. In either case it may momentarily not provide the right voltage or current on one of more of the lines to the motherboard which could cause any number of seemingly random errors.

Another power related possibility is that one of the many capacitors used for power smoothing on the motherboard has failed (this was a very common problem a few years ago, see http://en.wikipedia.org/wiki/Capacitor_plague for info including what to look for).

On XP's default of automatically rebooting on certain errors (that would otherwise result in a BSoD): this can be changed. Right-click "my computer", select "properties", goto the "advanced" tab, click the "settings" button in the "startup and recovery" section, and the dialogue box that comes up has a couple of options regarding what the kernel will do when it hits an error it can't cope with. This will mean you might get a useful error on-screen in place of a reboot (though if the problem is power related, the errors may not be helpful as the issue may affect different parts at different times - if you consistently get an error reported in the same module that might be a useful clue).

Answer 2

"Sounds like faulty memory - but memtest86+ says otherwise. A full test, taking over an hour, found no issues."

An hour on memtest is not a full test by any means, faults frequently show up after an hour. Run overnight to get something closer to a full test. And even then, it doesn't mean the memory is good, only that memtest found no fault in that time span.

Anyway, you don't need to run memtest. The problem has only happened since you installed the memory. Remove the new memory and see if the problem goes away.

If it runs fine without the new memory in it, then try swapping the new memory for the old and see what happens. If the problem returns, put back your old memory, and test the new memory a stick at a time with Memtest (at least 8 hours). If one tests good and one bad, try using only the good one. If both test good...well, you still can't use them.

It could be the memory is not at fault, but the extra load on the psu when you install the extra memory is making it sick. Unlikely, but i suppose not impossible.

Answer 3

As a source of random restarts I check HEAT, DISK, POWER and MEMORY in that order. However, random freezes are rarely (in my experience) power related so memory moves up one. As a source of restarts, disk is not a common one BUT disk errors are, in general, far more common than memory errors (again, in my experience) so I check that first.