Microsoft research on hardware failures.

[StrangerGuy]

How many of these unstable CPUs come from laptops with inherently poor cooling systems I wonder? Or from CPUs that are buried under an inch of dust?

 

[Bill Brasky]

How many of these unstable CPUs come from laptops with inherently poor cooling systems I wonder? Or from CPUs that are buried under an inch of dust?

Actually, laptops were significantly more stable than desktops. I think they were 60% less likely to have a fatal error. Your point can still be applied to desktops, though. It probably has more to do with the more extreme overclocking done a desktop vs laptops.

 

[Zxian]

What's funny is that they definitely didn't write that paper in Word.

 

[BonzaiDuck]

I scanned through the study, reading specific sections with a little less hurry.

They didn't create a "lab" of overclocked systems; they evaluated overclocked systems that were "out there" with data accessible for study.

So there are two obvious sources of errors -- probably already mentioned by posters here -- those reported during the tedious and painful overclocking process, and those due to careless over-clocking.

Or perhaps a third source -- errors that were statistically more likely to occur because OC'd systems were being run at the edge of or beyond the intended voltage and speed specs.

I just had my first "error" that caused a complete reset of the computer. Coincidentally -- it occurred today, and after more than a year of flawless operation. There was a message on reboot pointing to my UPS battery -- an APC with the PowerChute communication software.

The machine had gone into hibernation because -- for some reason -- it had been "seen" as running on battery longer than 3 minutes. [I remember creating this setting.] But it really hadn't been on battery; the battery showed to be charged 100% with 31 minutes of operation available.

Somehow, this seemed more like a software error or a hardware error connected to the APC. And the system didn't "BSOD" or crash; it recovered from hibernation with an Anandtech post-in-the-making, which I was able to then submit successfully without problem.

 

[Ferzerp]

You make a good point. When finding a stable OC, many, many crash dumps are created, and if you upload those, it sure looks like OC'ed systems are totally unreliable.