Merry Christmas and a prime! (M50 related)

[kriesel]

Originally Posted by Dubslow

I doubt that DC will ever be obsolete, even by PRPs. We simply cannot rule out unspeculated kinds of problems. It will be a major boon to nearly kill the 4% error rate, but DCing will live on.

You're probably right that DC will always be around. The creation of a PRP-DC work type and PRP-CF-DC work type are consistent with that.

I think we're already seeing, with the considerably lesser effort in cpu-hours going toward DC than first-time checks, that DC is not highly valued by many participants. As more reliable code gets widely adopted, that's likely to become more so. (Either LL with Jacobi, or PRP with Gerbicz, or any future error-check innovations.)

It's great that some of this has also been implemented for some GPUs. PRP in GPUOwl is the only one I know of currently. I hope its availability spreads to more software (CUDALucas, clLucas) and more device types (especially NVIDIA).

Unknown issues might be an argument for pairing PRP and LL on an exponent, either routinely, or as occasional pseudorandom spot check samples. Software bugs are less likely to affect both, if the parentage of the software is very separate. The more code is reused from one to the other, the less benefit of code independence there is. What I'm thinking of here is, separately from assignment of DC at the DC wavefront, ongoingly assigning some paired checks to different users, with different software, on different hardware types, at or ahead of the typical first-time assignments, as a form of continuous scouting for possible issues.

There have been double checks performed in every million interval through 106 million. I have been running DC on exponents to fill the gaps and have an exponent in 107 million running DC currently. If there is an issue with some fft length, it would be good to smoke that out early if possible, so there's time to debug and fix it, before the hordes of primenet-dispatched systems reach that fft length for routine assignments.

[ewmayer]

Originally Posted by kriesel

Widespread adoption of PRP could provide a boost to throughput, by making DC unnecessary for future exponents.

I doubt we'll ever obviate the need for DC ... the PRP + error-check will hopefully simply reduce the need for running more than the initial pair of tests.

[ixfd64]

Our friends at distributed.net just gave us another shout-out!

https://facebook.com/dnetc/posts/1903264189701370

[clowns789]

Also saw this advertised as a sponsored post by FedEx on Facebook...

http://about.van.fedex.com/blog/prime/

[ixfd64]

Originally Posted by clowns789

Also saw this advertised as a sponsored post by FedEx on Facebook...

http://about.van.fedex.com/blog/prime/

That is really cool. I hope this will encourage more companies to allow employees to run DC software on work computers.

[Spherical Cow]

Just in case it's not already noted- the new prime made it in to the journal Science. At the top of their News section. Prestigious placement in a prestigious journal. Though GIMPS isn't named specifically, it is mentioned as "a consortium of volunteers".

Norm

[ewmayer]

Originally Posted by Spherical Cow

Just in case it's not already noted- the new prime made it in to the journal Science. At the top of their News section. Prestigious placement in a prestigious journal. Though GIMPS isn't named specifically, it is mentioned as "a consortium of volunteers".

Norm

Hey, despite the blurb-brevity at least they properly got the "2 to the 77232917" part right, lots of stories I saw rendered that as 277232917. Speaking of said exponent, it's fun to play with various numeric bases to see if the exponent has an easy-to-remember digit pattern in some base. Using the obvious alphabetic extension [G = 16, H = 17, etc] to base-16 notation and restricting myself to bases <= 36 (beyond whch we run ot of letters and have to get creative to accommodate things],

23: BMMGLL
25: 7MHMGH

Note: I stuck to simple repeating-digit-style patterns, didn't look for patterns based on the alphabetic digits forming a word or license-plate-style abbreviation - I expect there is still fun remaining to be had there. Going a bit beyond base-36 and using base-10 numeric representation of the individual 'digits' w.r.to the base in question, a few curios:

54: 09 04 25 49 11, the leading 4 of which are perfect squares;
56: 07 47 43 47 13, all 5 digits are primes == 3 or 7 (mod 10).

[ATH]

GMP can do up to base 62, though it is a bit confusing.

For bases <=36 letters a-z corresponds to values 10-35 and are output in lower case.

For bases 37 to 62 suddenly upper case A-Z means values 10-35 and lower case a-z means values 36-61.



For 77232917:

Base22: ELF633
Which is apparantly an electric fan heater:
http://www.frico.se/en/products/Fan-...33-fan-heater/

Base52: ATEOL
which is a high blood pressure medication:
https://www.providentpharma.com/prod...ressure/Ateol/

[LaurV]

Up to now, the Fedex article seems the best (including the Gimps official release ) - we can give objective explanations why...

[ewmayer]

Just occurred to me that the above-copied Science blurb failed to include a link to GIMPS - yeah, I know you had to make it brief, but c'mon, guys...

[jasonp]

Maybe they figured everyone would google for 'gimps' and be horrified.