Possible garbage result in M84944207
 

I got the message (see attachment) in prime95. Does it mean the result is garbage?

[QUOTE=chuenkit;522669]I got the message (see attachment) in prime95. Does it mean the result is garbage?[/QUOTE]

Maybe. I guess you have a 50-50 chance of turning in a good result. You're this far in, I'd complete the test.

Get version 29.8. Change your work preference to what-makes-sense or first-time PRP. Error checking during a PRP test will catch every hardware error and repair the fault. Also, if you have any more LL tests queued up, unreserve them.

[QUOTE=Prime95;522671]Maybe. I guess you have a 50-50 chance of turning in a good result. You're this far in, I'd complete the test.

Get version 29.8. Change your work preference to what-makes-sense or first-time PRP. Error checking during a PRP test will catch every hardware error and repair the fault. Also, if you have any more LL tests queued up, unreserve them.[/QUOTE]

Noted. I think I will move to work on double-checking.

[QUOTE=chuenkit;522673]Noted. I think I will move to work on double-checking.[/QUOTE]

Flaky hardware can best contribute with first time PRP testing. All hardware errors are caught and corrected.

Double-checking will run LL tests which may or may not catch hardware errors.

[QUOTE=Prime95;522674]Flaky hardware can best contribute with first time PRP testing. All hardware errors are caught and corrected.

Double-checking will run LL tests which may or may not catch hardware errors.[/QUOTE]
Hmm, I see both Double check LL and Double check PRP choices in both the manual assignment page and in prime95 v29.8b3 worker window type of work to get. Have a PRP-DC in progress in prime95.

[QUOTE=kriesel;522693]Hmm, I see both Double check LL and Double check PRP choices in both the manual assignment page and in prime95 v29.8b3 worker window type of work to get. Have a PRP-DC in progress in prime95.[/QUOTE]

Yes, PRP-DC is also a suitable work preference. However, you will be testing numbers of approximately the same size. First time PRP only started a about a year ago with exponents in the 80 millions.

I think it's great any time someone is willing to chip away at the big backlog of LL-DC. And also when anyone is willing to contribute some PRP-DC, which are relatively scarce, so there's not much sample size for seeing any possible escapes of errors from the excellent Gerbicz Error Check and code surrounding it. As I recall, George did a code review a while back and found some possible cases, since fixed. Gpuowl PRP has been available for about 2 years and was running 4M fft around 75M up to 78M as I recall.

[QUOTE=Prime95;522674]Flaky hardware can best contribute with first time PRP testing. All hardware errors are caught and corrected.

Double-checking will run LL tests which may or may not catch hardware errors.[/QUOTE]

Maybe flaky hardware can contribute most by doing PRP double checks, to help verify that the new improved error checking really is bulletproof.

[QUOTE=GP2;522768]Maybe flaky hardware can contribute most by doing PRP double checks, to help verify that the new improved error checking really is bulletproof.[/QUOTE]

I agree with all that has been said. My post came across as poo-poo-ing PRP-DC work. That was a mistake. I was trying to give the user more info for making his choice.

We all agree that PRP or PRP-DC work is the best choice.

1) First-time PRP work will get exponents around 89 million. Best chance of finding a new prime.
2) Double-check PRP work will get exponents around 80 million, well ahead of the LL-DC wavefront. Poor chance of finding a new prime. Work is especially useful to verify the quality of early PRP error checking implementations.

[QUOTE=chuenkit;522669]I got the message (see attachment) in prime95. Does it mean the result is garbage?[/QUOTE]

Running a double check on this, eta 4 days. :smile:

Successful DC. Horray!

(This links to exponent status of M84944207)
[url]https://www.mersenne.org/report_exponent/?exp_lo=84944207&full=1[/url]

And with that completed, I believe all exponents below 85 million have been checked at least once.

Double Hooray!! :max: