View Single Post
Old 2013-01-26, 15:02   #33
P90 years forever!
Prime95's Avatar
Aug 2002
Yeehaw, FL

37·193 Posts

Originally Posted by henryzz View Post
I though we needed 2 non-x86 runs last time. The gpu run is one. Do we need another?
I think I've only ever *required* one different-hardware-different-software run. More just adds weight to the claim.

I'd be willing to reduce that to different software on the same hardware type. Personally I'd accept Ernst's Mlucas run on an x86 machine because as an LL / FFT programmer I know it would be virtually impossible for a previously unknown CPU bug to cause two different LL programs to fail with a false prime report.

The whole concept of "when is a new prime considered independently verified" is kind of interesting. How important is the reputation of the verifier? Do GIMPS' claims require less independent verification because of its past track record (or equivalently does a unknown newcomer require a more thorough verification)? How much does the open source nature of the GIMPS and Mlucas software help a claim? How important is the claimant's ability to describe the process / algorithms used?

I know I wouldn't trust a new claim by any of the cranks in the Misc Math subforum if it was independently verified by Joe Blow, John Doe and 10 other no-name individuals -- at least not until I reran the test myself.

Last fiddled with by Prime95 on 2013-01-26 at 15:03
Prime95 is online now   Reply With Quote