Will testing stop the moment Gerbicz-PRP fails?

[011hoot]

Am running a Gerbicz error-checking PRP test of a Category 4 exponent. This is my first time.

Will the PRP test stop the moment a test fails?

Also, where does the 1/840924 probability of success come from?

Thanks,
011hoot

[Prime95]

Depends what you mean by "fails".

If Gerbicz error checking fails (detects an error) the PRP test backtracks a bit and proceeds onward.
If the PRP test fails (detects a not-prime result) there will be a very short proof generation step then you'll move onto the next exponent to PRP test.

The probability comes from the exponent, how much trial factoring has been done, and a rough adjustment for P-1 factoring not finding a factor. Displaying 6 digits of precision is a bit silly, more accurate would be "about 1 in 841000".

Welcome aboard!

[PhilF]

Welcome!

You might be interpreting the word "fail" incorrectly. Do you mean will the test stop if the Gerbicz error check fails, as in there is an error in the test? Or do you mean will the test stop if the Gerbicz test fails, meaning the number is not prime?

If it is the latter, then no, a PRP test must be run all the way to the end before it can determine whether a number is a probable prime or composite.

[Zhangrc]

Originally Posted by Prime95

If Gerbicz error checking fails (detects an error) the PRP test backtracks a bit and proceeds onward.

I've never encountered it until yesterday. When Gerbitz checking failed, I was frightened. My exponent, M109529873, was 96% done. So I decided to finish it anyway, instead of starting from scratch. Let's see if the cert will be successful.

[Apr 27 12:22] Iteration: 107000000 / 109529873 [97.69%], ms/iter: 10.087, ETA: 07:05:18
[Apr 27 12:22] ERROR: Comparing Gerbicz checksum values failed.  Rolling back to iteration 106000000.
[Apr 27 12:22] Continuing from last save file.
[Apr 27 12:22] Setting affinity to run helper thread 1 on CPU core #2
[Apr 27 12:22] Setting affinity to run helper thread 2 on CPU core #3
[Apr 27 12:22] Setting affinity to run helper thread 3 on CPU core #4
[Apr 27 12:22] Trying backup intermediate file: p109529873.bu
[Apr 27 12:22] Trying backup intermediate file: p109529873.bu2
[Apr 27 12:22] Trying backup intermediate file: p109529873.bu3
[Apr 27 12:22] Resuming Gerbicz error-checking PRP test of M109529873 using FMA3 FFT length 6M, Pass1=1536, Pass2=4K, clm=1, 4 threads
[Apr 27 12:22] PRP proof using power=8 and 64-bit hash size.
[Apr 27 12:22] Proof requires 3.5GB of temporary disk space and uploading a 123MB proof file.
[Apr 27 12:22] Iteration: 106000001 / 109529873 [96.77%].
[Apr 27 12:22] Hardware errors have occurred during the test!
[Apr 27 12:22] 1 Gerbicz/double-check error.
[Apr 27 12:22] Confidence in final result is excellent.
[Apr 27 12:24] Iteration: 106010000 / 109529873 [96.78%], ms/iter: 10.275, ETA: 10:02:45
[Apr 27 12:24] Hardware errors have occurred during the test!
[Apr 27 12:24] 1 Gerbicz/double-check error.
[Apr 27 12:24] Confidence in final result is excellent.
[Apr 27 12:26] Iteration: 106020000 / 109529873 [96.79%], ms/iter: 10.253, ETA: 09:59:47
[Apr 27 12:26] Hardware errors have occurred during the test!
[Apr 27 12:26] 1 Gerbicz/double-check error.
[Apr 27 12:26] Confidence in final result is excellent.
[Apr 27 12:27] Iteration: 106030000 / 109529873 [96.80%], ms/iter: 10.298, ETA: 10:00:40
[Apr 27 12:27] Hardware errors have occurred during the test!
[Apr 27 12:27] 1 Gerbicz/double-check error.
[Apr 27 12:27] Confidence in final result is excellent.

That's the first error my computer has made in the past 2 years. It's now running out of maintenance period.

[kriesel]

Not to worry. When a GEC error is detected, prime95 or gpuowl retreat to the last known good saved residue and try again from there, and generally the error does not reoccur. I've had runs on unreliable hardware that had successful proofs and CERTs, after accumulating error counts in the hundreds. There have been cases where I found it necessary to switch to a longer fft size in gpuowl to get through an iteration range.

An old Athlon system I bought for $15 became so unreliable, that a PRP/GEC could no longer make progress, so the motherboard went to recycling. That's how good the GEC is.