Should I abandon this test?

JuanTutors · 2021-10-14, 12:10

I am doing a PRP test on a 10^8 digit Mersenne for which I am only at 3.79% of the way done. It has already had 2 Gerbicz/double check errors. It won't finish until maybe September 2022. Should I just abandon it/unreserved it and start a new test? FYI I have already stopped the test to do a DC assignment that passed.

Viliam Furik · 2021-10-14, 18:02

Quote:

Originally Posted by JuanTutors

I am doing a PRP test on a 10^8 digit Mersenne for which I am only at 3.79% of the way done. It has already had 2 Gerbicz/double check errors. It won't finish until maybe September 2022. Should I just abandon it/unreserved it and start a new test? FYI I have already stopped the test to do a DC assignment that passed.

If in the event of abandoning this test, and starting another 100M digit one, I strongly recommend not to do this. However, if you start with some smaller number to test, i.e. under 150M exponent, I would recommend doing this.

kruoli · 2021-10-14, 18:08

You should also pay attention on the errors: Doing DC instead or LL likely will soon cause a bad result! You could do "normal" PRP with proofs or even LL-DC by doing a PRP with proof; there is also PRP-CF which has proofs.

JuanTutors · 2021-10-14, 18:35

Quote:

Originally Posted by kruoli

You should also pay attention on the errors: Doing DC instead or LL likely will soon cause a bad result! You could do "normal" PRP with proofs or even LL-DC by doing a PRP with proof; there is also PRP-CF which has proofs.

I am doing a first time PRP test.

JuanTutors · 2021-10-21, 12:30

Update to this post: I did decide to switch exponents since I was so early into the test. The thing is, with my new exponent, I am only 2.15% of the way through the newest test and I am already at 3 Gerbicz errors in. Should I quit this machine? Should I use it to just constantly P-1 on my current assignments? Should I keep going with this PRP test?

kruoli · 2021-10-21, 12:45

If your machine is that problematic, PRP-CF is an option, since its assignments are so short that they should come through nicely.

P-1 is risky if the machine is unstable, factors are likely missed. If a normal P-1 would turn up with a factor, a single problem in the computation would prevent that factor from coming up. With you machine it is thus likely that it might never find a factor.

Needless to say, you should try to investigate why the errors occur. Unstable memory? Over-/underclock? Maybe try the stress test; are small FFTs stable? Are big FFTs stable?

paulunderwood · 2021-10-21, 12:50

Quote:

Originally Posted by JuanTutors

Update to this post: I did decide to switch exponents since I was so early into the test. The thing is, with my new exponent, I am only 2.15% of the way through the newest test and I am already at 3 Gerbicz errors in. Should I quit this machine? Should I use it to just constantly P-1 on my current assignments? Should I keep going with this PRP test?

Find out why you are getting Gerbicz errors. In the first place check your thermals with some temperature monitoring software. Other than this voltage and timings are crucial. What is the specifiction of your computer system?

JuanTutors · 2021-10-22, 17:13

Quote:

Originally Posted by kruoli

Maybe try the stress test; are small FFTs stable? Are big FFTs stable?

Will have to do this.

Quote:

Originally Posted by paulunderwood

Find out why you are getting Gerbicz errors. In the first place check your thermals with some temperature monitoring software. Other than this voltage and timings are crucial. What is the specifiction of your computer system?

Average temp hovering between 78C and 80C. Intel Core i5 6500 (Skylake) 3.20 GHz processor (though core temp says 3292.74MHz). TDP 65.0 Watts. Power around 40.0 Watts.

paulunderwood · 2021-10-22, 17:36

Quote:

Originally Posted by JuanTutors

Average temp hovering between 78C and 80C. Intel Core i5 6500 (Skylake) 3.20 GHz processor (though core temp says 3292.74MHz). TDP 65.0 Watts. Power around 40.0 Watts.

On warmer days this might be a problem. Is it a stock Intel cooler? Would you consider getting an all-in-one water cooler?

I have run Intel chips at these temperatures, but if heat builds in the case it might overheat other components such as RAM or bridge chips.

JuanTutors · 2021-10-22, 18:01

Quote:

Originally Posted by paulunderwood

On warmer days this might be a problem. Is it a stock Intel cooler? Would you consider getting an all-in-one water cooler?

I have run Intel chips at these temperatures, but if heat builds in the case it might overheat other components such as RAM or bridge chips.

It's a work PC so that's not an option but will throttling do it? I am definitely willing to restart this test if that might work.

Also an update, I just tried switching BACK to the number that I had stopped working on just last week and there was an error reading ALL the intermediate files. I don't even have a viable explanation for that.

paulunderwood · 2021-10-22, 18:06

Quote:

Originally Posted by JuanTutors

It's a work PC so that's not an option but will throttling do it? I am definitely willing to restart this test if that might work.

Also an update, I just tried switching BACK to the number that I had stopped working on just last week and there was an error reading ALL the intermediate files. I don't even have a viable explanation for that.

There is no need to restart your number (after throttling). You can be very confident with any number of Gerbicz errors.

2021-10-14, 12:10	#1
JuanTutors "Juan Tutors" Mar 2004 569 Posts	Should I abandon this test? I am doing a PRP test on a 10^8 digit Mersenne for which I am only at 3.79% of the way done. It has already had 2 Gerbicz/double check errors. It won't finish until maybe September 2022. Should I just abandon it/unreserved it and start a new test? FYI I have already stopped the test to do a DC assignment that passed. Last fiddled with by JuanTutors on 2021-10-14 at 12:11

2021-10-21, 12:45	#6
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 1,087 Posts	If your machine is that problematic, PRP-CF is an option, since its assignments are so short that they should come through nicely. P-1 is risky if the machine is unstable, factors are likely missed. If a normal P-1 would turn up with a factor, a single problem in the computation would prevent that factor from coming up. With you machine it is thus likely that it might never find a factor. Needless to say, you should try to investigate why the errors occur. Unstable memory? Over-/underclock? Maybe try the stress test; are small FFTs stable? Are big FFTs stable? Last fiddled with by kruoli on 2021-10-21 at 12:45 Reason: Punctation.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
I found the primality test, there seems to be no composite numbers that pass the test	sweety439	sweety439	7	2020-02-11 19:49
Modifying the Lucas Lehmer Primality Test into a fast test of nothing	Trilo	Miscellaneous Math	25	2018-03-11 23:20
Double check LL test faster than first run test	lidocorc	Software	3	2008-12-03 15:12
Will the torture test, test ALL available memory?	swinster	Software	2	2007-12-01 17:54
A primality test for Fermat numbers faster than Pépin's test ?	T.Rex	Math	0	2004-10-26 21:37

2021-10-14, 18:08	#3
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 1,087 Posts	You should also pay attention on the errors: Doing DC instead or LL likely will soon cause a bad result! You could do "normal" PRP with proofs or even LL-DC by doing a PRP with proof; there is also PRP-CF which has proofs.

2021-10-21, 12:30	#5
JuanTutors "Juan Tutors" Mar 2004 569 Posts	Update to this post: I did decide to switch exponents since I was so early into the test. The thing is, with my new exponent, I am only 2.15% of the way through the newest test and I am already at 3 Gerbicz errors in. Should I quit this machine? Should I use it to just constantly P-1 on my current assignments? Should I keep going with this PRP test?