Error message

[chesswood]

The following gets logged on stdout every 10000 iterations:

Possible hardware errors have occurred during the test! 1 ROUNDOFF > 0.4
Confidence in final result is fair.

What am I supposed to do? Buy a new computer?

[VBCurtis]

Are you overclocking? If so, turn down the overclock.

Have you removed the dust and dirt from your machine in the last year or two? It might be slightly overheating because the dust is stifling airflow and insulating the surfaces it rests upon.

If those aren't culprits, you might have found the first sign of failing RAM on your machine. Or, perhaps it was a one-time fluke and the machine is mostly-fine. A single error report isn't a conclusive sign of much of anything, but it is a nice reminder to check on the above items.

[Thecmaster]

Originally Posted by chesswood

The following gets logged on stdout every 10000 iterations:

Possible hardware errors have occurred during the test! 1 ROUNDOFF > 0.4
Confidence in final result is fair.

What am I supposed to do? Buy a new computer?

I got the same after OC and tuned back OC and I used a backup that was good to go back before error accrued. No more errors after that.

If you don't have a good backup you can restart.

You can pause LL and do a torque test for 12-24 hours after tuning back OC if you have done that. Also do a memory test in prime95.

[LaurV]

This means that sometimes in the past during this particular test, a hardware error occurred. It does NOT mean that you are getting continuously errors, once (or more) at every 10k iterations. The information is stored in the checkpoint files and it is "reminded" to you to be sure you see it and decide what to do with the test.

That means nothing for a single occurrence. If the issue repeats, if it appears often, etc, then it is different.

If the test is not very advanced (like losing weeks, but only days or hours), I would stop P95, delete checkpoint files and restart.

If the problem repeats, do what other guys said (reduce OC. remove dust clogs, etc).


Edit, OTOH, your exponent may be at a FFT borderline, and just get a 0.4 roundoff. In this case, the error will repeat at exactly the same point when the test is restarted, assuming the shift is the same, and increasing the FFT manually or changing the shift will solve the problem. In fact, P95 will chose a random shift at every restart, if not otherwise forced.