ERROR: SUM(INPUTS) != SUM(OUTPUTS)

[ebx]

Hi,

I have seen a few errors in my log files at 2 different hosts, one at home and one in a rack room.

Two error types:

ERROR: ROUND OFF (0.4999008179) > 0.40

ERROR: SUM(INPUTS) != SUM(OUTPUTS), 6.671785364156798e+16 != 6.671785364161534e+16

Most are SUM. Both cases followed by

Possible hardware failure, consult the readme file.
Continuing from last save file.

Torture for hours didnt produce any info. They all started in a very hot day. I dont know one in rack room but my home CPU was over 60c.

How bad are they?

[Digital Concepts]

Overheating does cause errors in CPU and memory. George wrote here http://www.teamprimerib.com/gimps/viewtopic.php?t=49, that you have a 50% change of the exponent run being invalid.

Prime95 detected that something happened when numbers went out of spec. It kept reruning from the last known good checkpoint until numbers went back into spec. The problem is the overheat may have harmed work done before that checkpoint (but didn't know it yet) and this would then carry through the rest of the WU.

[ebx]

RATS! 56% is much too high. Will errors be reported to primenet so it could set a flag for those exponents?

Or offer an option that when we see such an error, we start from the beginning.

[garo]

Yes the errors are reported to primenet. Anyway, doublechecking ensures that only matching results are accepted.

If your CPU temp is back to normal and you have less than 50% of the exponent complete I'd say start over.

Stop Prime95.
Go into the P95 directory and look for two files named pxxxxxxx and qxxxxxx where the x's are numbers representing your exponent. Remove the two files and restart P95 and you should start over.

Hi, I got all these errors (and no others) over a period of a few days.

The errors seem to all be very small; within a reasonable floating-point epsilon.

I ran the torture test for a few hours without failures. My AMD Athlon 2700+ isn't overclocked. It's not hot here.

I'm disappointed; I had hoped to be able to help with the Mersenne prime search. But, I am going to discontinue running this computation.

If in fact these are epsilon errors and the results from my computer have a chance of being useful, could someone let me know?

Thanks!
Oliver Jones


Iteration: 4592344/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 1.276158324226189e+83 != 1.27615832422619e+83
Iteration: 4691536/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 2.917352435159693e+82 != 2.917352435159695e+82
Iteration: 6706828/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 6.538809292854271e+66 != 6.538809292854284e+66
Iteration: 6791001/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 6.548121686333023e+66 != 6.548121686333019e+66
Iteration: 6841061/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 1.113625838140894e+67 != 1.113625838140895e+67
Iteration: 7264636/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 1.70034274692042e+76 != 1.700342746920419e+76
Iteration: 7499587/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 1.446904841557651e+67 != 1.44690484155765e+67
Iteration: 7622131/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 6.479168211454303e+67 != 6.479168211454303e+67
Iteration: 7635629/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 2.365930387327026e+67 != 2.365930387327026e+67
Iteration: 8828490/22024853, ERROR: SUM(INPUTS) != SUM(OUTPUTS), 3.649538737865422e+83 != 3.649538737865421e+83

[nfortino]

Originally posted by OliverJones
If in fact these are epsilon errors and the results from my computer have a chance of being useful, could someone let me know?

Completed exponents with 10 sum(inputs)!=sum(outputs) historically have a 60% chance of being bad, and therefore you should not finish it. Try running Memtest86 to see if bad memory is the cause of your problem. If that comes clean, try running a few double checks, and when they complete check here (link at bottom of page) to check if they match the other result.