In a random look at recent work history I was surprised to find a couple of units with the big red WARNING status on them. That's something for me to look at separately. What I found interesting was that in looking through the results without such a warning, I saw the following:

Iter: 17819779/35148164, ERROR: ROUND OFF (0.403055214) > 0.4

In a random look at recent work history I was surprised to find a couple of units with the big red WARNING status on them. That's something for me to look at separately. What I found interesting was that in looking through the results without such a warning, I saw the following:

Iter: 17819779/35148164, ERROR: ROUND OFF (0.403055214) > 0.4

To me, this is another bad state to be in, yet it was successfully (silently) recovered in this specific case. My request is for this to also be flagged with a WARNING status. If I understand correctly, the current warning is only shown if there is a Gerbicz check failure, indicating a computational error that was detected. This would be a similar state.

Unit showing the above error:
http://www.primegrid.com/result.php?resultid=1205582261

The same machine did show Girbicz errors around the same time so in this particular example it is probably a real error. Presumably because it was detected and re-run successfully outside of the Girbicz check, that was not indicated.

That there might be false positives in the case of borderline FFT sizes didn't occur to me. Still, if there are real computation errors happening, I guess the chances are it'll also show up in a Girbicz error sooner or later.