PrimeGrid
Please visit donation page to help the project cover running costs for this month
1) Message boards : The Riesel Problem : K = 192971 prime found by Propper (Message 150373)
Posted 7 days ago by Profile Michael GoetzProject donor
But the point is that it is possible to generate the additional files, tarball them, and ship them off to anyone that wants to quickly verify the work. Yeah, it is a couple extra steps, but it seems insanely worthwhile to do.


But it has to have *already* been done for the Gap Numbers. If it hasn't (and there's no reason to believe it has been, and lots of reason to believe it hasn't), then we're back to square one: if anyone is going to test those Gap Numbers, then the tests Propper has already done on them are wasted.
2) Message boards : The Riesel Problem : K = 192971 prime found by Propper (Message 150370)
Posted 7 days ago by Profile Michael GoetzProject donor
Maybe Primegrid should think about testing the "Gap Numbers" in another way.

First, PG should continue the search ignoring the "Gap Numbers" and focus only on numbers without prime. And for sure, there will be other primes discovered outside PG.

When all remaining TRP will be found (not if ...), and I think it will probably be in a decade


I'm sure it will be done in "a" decade. But I doubt it will be the current decade, or the decade after that, or the decade after that...

I'd wager that it would be closer to a century (or centuries) than a decade, but I won't because it's unlikely I'll be around to collect. :)
3) Message boards : The Riesel Problem : K = 192971 prime found by Propper (Message 150369)
Posted 7 days ago by Profile Michael GoetzProject donor
Oddly enough, if LLR2 (and thus fast double checking) didn't exist, this wouldn't matter. The norm would be to run two tests on each number (a check and a double check). Propper could simply provide the residues for his tests on the Gap Numbers, and a second test could be run to confirm those residues. But with LL2, the norm is now just one test, and there's no way to confirm the Gap Number results without running two tests.

If he were to run and provide the output files from LLR2, wouldn't the same be achievable? Wouldn't we be able to generate just the quick verification task if we were given the correct inputs to generate it?


LLR2's does not operate in fast double check mode unless you specifically tell it to do so. Normal LLR/LLR2 command lines looks like this:

llr -d -t4 -q"12345*2^12345678+1"

Take a look at the command line on any LLR2 task. It's a lot different. It's not something you would do without explicitly trying to do so.

Furthermore, the "output files" are gigabytes upon gigabytes of proof files. I can't imagine he's running LLR2 in that mode, even less so storing those files, and these files would have to be stored somewhere until someone decided they wanted to test the Gap Numbers. It's those large proof files that let us do the fast double checks. With just the residues, a full double check is needed.
4) Message boards : The Riesel Problem : K = 192971 prime found by Propper (Message 150357)
Posted 7 days ago by Profile Michael GoetzProject donor
why are you saying that third-party searches are a waste of resources? If the found K stop searching, then this speeds up the search for the remaining ones. it is beneficial for all of us


Thank you for asking this. It is a great question.

For the sake of clarity, let's call the numbers between where PrimeGrid has searched and Propper's Prime as the "Gap Numbers".

If all we're comcerned about is finding a prime for each K in order to prove the conjecture, then you would be correct. While it is true that this is the primary concern, it's not the only concern. There is also some value in determining that all the smaller numbers (including the Gap Numbers) are composite so that an accurate accounting of the distribution of primes can be determined. For this reason, PrimeGrid has historically also made sure to fully check and double check all the numbers below each prime.

Until now, it was our intention to continue doing this, but we made the decision to not search the Gap Numbers for primes found outside of PrimeGrid.

However, since there is value in proving all those numbers are composite, I'm assuming that someone will eventually go back and test the Gap Numbers. They would need to be checked again because nobody can prove the accuracy of Propper's tests on the Gap Numbers.

Oddly enough, if LLR2 (and thus fast double checking) didn't exist, this wouldn't matter. The norm would be to run two tests on each number (a check and a double check). Propper could simply provide the residues for his tests on the Gap Numbers, and a second test could be run to confirm those residues. But with LL2, the norm is now just one test, and there's no way to confirm the Gap Number results without running two tests.

The only way to confirm the Gap Numbers is to run two tests, Propper's already completed test and a hypothetical future second test. If they had been done with LLR2's fast doublechecking, only one test would have been needed.

That's the waste of resources.

That doesn't stop us from being happy that the primes were found, moving us closer to solving the conjecture.
5) Message boards : Extended Sierpinski Problem : Main task - Proof task. (Message 150204)
Posted 21 days ago by Profile Michael GoetzProject donor
Hello. I have some questions.

- Can someone tell me how the "main task" and "proof task" are distributed? I'm still learning.

- Can I choose what kind of homework to do? or does the system choose for itself?

- In case of being able to "choose". How can I do it?

Thank you very much for the help and I regret my English, I am doing it through an automatic translator.


It's all automatic, and no you can't choose.

After you complete a main task and return the result, the server does some magic and creates a proof task. The proof task (about 1% the size of the main task) is put in the queue and sent out to a computer, and when it's returned, assuming both computers produced the correct result, both the main task and the proof task are granted credit.

It's similar to normal double checking where the same tasks are sent to two different computers, except:

1) The second task isn't sent until the first task is returned.
2) The second task is much smaller, and gets proportionally less credit.
3) It's alright if the same computer does both tasks.
6) Message boards : Number crunching : Yuri's Night Challenge (Message 150101)
Posted 32 days ago by Profile Michael GoetzProject donor
btw How much in is WW currently? How much of a dent was this challenge able to make?


The leading edge is now at 22.33%. It was at 19.65% at the start of the challenge.
7) Message boards : Number crunching : Yuri's Night Challenge (Message 150062)
Posted 34 days ago by Profile Michael GoetzProject donor
Should I be running this on CPU too?


Yes. It won't add a LOT of extra tasks, but every little bit counts.


What's the optimal number of threads?


All of them. Including hyperthreads. Seriously.

This is very slow on the CPU, and the challenge is only three days long, (and now there's really only one day left), so you need the tasks to finish as quickly as possible. I'm using three computers right now which are 8c/16t, 4c/4t, and 2c/4t, and I'm running WW on those CPUs at -t16, -t4, and -t4, respectively. The run times for those machines in that configuration are approximately 6 hours, 24 hours, and 72 hours. Yes, that last machine will barely complete one task. Maybe. it will be close.

EDIT: The three CPUs are AMD 3700X, Intel i5-4670K, and Intel i3-330M. No, I'm not bothering to reserve any threads for the GPU. WW hardly uses any CPU time, and keeping all threads crunching doesn't seem to impact the GPU speed, at least not on my 3700X/GTX 1060 combination.

EDIT2: As of right now, it looks like that last machine will finish its one and only task with about 40 minutes to spare. It started the task 7 seconds after the challenge started.
8) Message boards : Fermat Divisor Search : Fermat Divisor Search (Message 150012)
Posted 36 days ago by Profile Michael GoetzProject donor
Can it be removed from the pg prefs page?


When I get around to it, yes. I won't touch it until after the challenge is over, however.
9) Message boards : Fermat Divisor Search : Fermat Divisor Search (Message 150010)
Posted 36 days ago by Profile Michael GoetzProject donor
Hello!

As a follow up to dannyridel's question, what project will become the new Focus project?

May I suggest SR5, relative short running time a lot of candidates and sooner (or later!) we can prove a conjecture the better!😁

With regards,

Hans Sveen
Oslo


I've removed the focus project designation from DIV.

At this time, there's no particular project that we want people to focus on, so there not going to be a focus project for now.
10) Message boards : Number crunching : Yuri's Night Challenge (Message 150009)
Posted 36 days ago by Profile Michael GoetzProject donor
how much work remains on this task. what are the approximate deadlines for the completion of the subproject?


We're at 19.65 percent.

The leading edge is 3625129000000000000 and assuming the software isn't changed, it will stop working at 18446744073709551616.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 5.37, 5.27, 4.92
Generated 18 May 2021 | 5:22:58 UTC