It's not the power supply. I just swapped it with another computer's (a fairly new, 450W one) and the same thing happened.

Next is a memory diagnostic.

[Update at 12:50 PM EDT]

Memtest86 found five errors in the first pass, on tests 6 and 7. The problems with the Athlon supposedly show up on test 5 or 8, so it's probably not a speed problem. I'll let it run another pass and see if it sees the same thing the second time through.

[Update at 1:17 PM]

Only one of the errors repeated, on test 6, but it did repeat.

[Update at 2 PM]

I let the machine reboot into Windows after the memory test, and it's now been up for over half an hour, which is a record for the last couple days. I don't know why running a memory test would have that effect, but the problem seems to have been (at least) mitigated, at least temporarily.

[Update at 8:40 PM EDT]

OK, the machines is now up for over six hours, with no glitches. I'm posting from it. Go figure.

No, thanks, I know this isn't the end of the story.

[Friday morning update]

The machine is still alive and healthy this morning.

How old is the box? I haven't had computer problems like that since I began replacing critical components every 2.5 years. Of course, that's not cheap, but it's been awhile since I seen a head ache like your having.

I have had those head aches though, and I have lots of empathy. I've had a power supply blow and take out the CPU and a memory stick.

The case is several years old. The power supply is (now) a few months old, and barely used in that time. The processor is a year or two old, as is the fan. The MB is a few months old. The memory is two or three years old.

I'd bet on a problem with the RAM. Memory problems can come or go with temperature, but tend to get worse over time. You run the danger of data corruption. First, I'd reseat the modules (while carefully grounded). If you have multiple RAM modules, and depending on the architecture, you might start pulling modules and retest. Failing that, I'd buy at least one replacement module and retest.

Have you had many thunderstorms lately? Even with surge protectors, dangerous current can get through sometimes and fry chips.

Did you build this system yourself or did you buy it OEM?

If built, Did you install the chipset drivers from the CD that came with the motherboard? What kind of Motherboard is it? If a built system then its best to use the drivers that came with the mainboard. My experience is that the latest greatest drivers from the manufacturer of the chipset device usually just end up breaking the system. The supplier of the mainboard tested the package with the driver that was shipped and that is all they will gaurantee to work.

If you go into the BIOS, assuming this is a built system, then you should have several different CMOS profiles to choose from. See if there is one that will load fail-safe defaults.

Also, if you check under the advanced chipset features there will be a section for something like 'Memory Frequency' By default this maybe set as 'By SPD' which tells the BIOS to use the memory timings encoded on the memory modules themselves. Usually AUTO is the best selection to choose because then the system will actually check with the processor to see what it can handle instead of imposing what the Memory modules think the processor can handle. Normally 'BY SPD' timings are the best to use from a performance standpoint but one has to take great care in selecting a processor/memory combo that is compatible with that setting. OEM systems won't give you that level of control by default. There are progrms though that will give you more information about the BIOS settings then what is displayed by an OEM bios.

The reason I was asking about the chipset drivers is because there is a logic memory controller built into the chipset that interfaces with your system RAM. There are memory address remapping tables coded into the memory controller structure that set address ranges for the O/S to interface with. MS Windows can hold quite a bit of control over the lower level hardware functions and for the sake bootup performance will tend to use the last set of known good values when loading Kernel system DLL's into the same memory address spaces over and over again. When you booted up into a difference O/S from the memtest boot disk you might have given your hardware a chance to relearn in essence what were a valid set of memory address ranges. Then, when booting up Windows again it would have loaded system dll's into the cooresponding valid address ranges.

Sounds good anyways :)

I built it myself, and no, I didn't use the CD that came with the Mobo--I just reinstalled Windows over the old installation.