PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Generalized Fermat Prime Search : DO YOU FEEL LUCKY? (Message 145351)
Posted 6 days ago by Kellen
Too bad we're not manual sieving n=22. We'd be removing factors 500 times faster than running GFN.
What is this about?


GFN-22 tasks take a very long time to run, and at the current sieve depth we can eliminate about 500 candidates by sieving in the time it takes to test one. Sieving for these is done by manually running software and uploading the factor files, but it has been suspended for a while, as Jim was the person taking care of this.

It will likely return in the future, but I think that there is so much excellent other stuff going on with PrimeGrid at the moment, that it is not as high on the priority list. There has also been a new application made by Yves since the sieving was closed that runs significantly faster for the smaller GFNs and when things start back up again it would make the most sense to use that program instead.

With LLR2 and the fast-DC, plus all of the other things that have been happening in the background, restarting manual sieving has taken a back seat. It will happen one day I'm sure. I am also hoping that gfsieve will receive BOINC integration so that manual sieving becomes a sub-project, but we will have to see about that I suppose ;)
2) Message boards : Storm message board (Message 145350)
Posted 6 days ago by Kellen
How seriously is everyone talking this latest challenge? :-)


My brains are still in full-on melt-mode and oozing out my ears looking for some of the xGFNs for Jeppe, so I'm out on this one. I was planning on being on it full-force, but insanity reigns supreme in the land of the xGFN at the moment XD
3) Message boards : Storm message board (Message 145050)
Posted 18 days ago by Kellen
I think you were right on the money: One had FFT as 2M, the other had FFT as 1920K.
The next two tasks that started are behaving in the same way, so the server must be sending out a mix of these at the moment...

Using zero-padded FMA3 FFT length 2M, Pass1=2K, Pass2=1K, clm=1, 6 threads, a = 3, L2 = 395*370, M = 146150

Using zero-padded FMA3 FFT length 1920K, Pass1=384, Pass2=5K, clm=2, 6 threads, a = 5, L2 = 437*332, M = 145084


Perfect! Glad this was actually the issue :)

With most projects there is a very narrow range of n values where you will get a semi-random FFT size while transitioning from one FFT size to the next, but for Cullen and Woodall that range is somewhat larger than normal; maybe even up to about 100,000. One of the things that affects it is odd vs even k and n for these. If k is even, you can divide it by two and increase n by one, which results in a lower FFT size. If you check through your results you will probably notice that all of the 2M FFT tasks are odd k and n and all of the 1920K FFT tasks are even k and n.

The same principle holds true for other projects, but because k is small in the PPS projects, the transition range is also small, and for a project like SGS the transition range would be small because n is small, but for a project like Cullen or Woodall where both k and n are large, this range is big :)

Edit: Congrats on prime #2!!
4) Message boards : Storm message board (Message 145036)
Posted 18 days ago by Kellen
It is possible that we've reached an FFT transition, at which time the larger FFT task would run slower. Your earlier measurements for speed probably involved two 1920K FFT length tasks, but as soon as the FFT size reaches 2048K (2M), two tasks will not fit in cache anymore. One 1920K task and one 2048K task would fit, so perhaps we are right at that threshold.

We will certainly reach it during the upcoming challenge where every task will be 2048K FFT length, at which time the performance difference between running two tasks and running 1 task will become much larger.

This would be my guess anyway, but you'll have to wait til those tasks complete to see the FFT sizes :)
5) Message boards : Storm message board (Message 145031)
Posted 18 days ago by Kellen
Woohoo! Congrats!

I don't know if you've been following the PrimeGrid chat where Rytis (founder of PrimeGrid) is looking into the ability to get your primes printed as a book, but this would be a good one for that :)

Good luck with future primes!
6) Message boards : Storm message board (Message 144916)
Posted 24 days ago by Kellen
Still plugging away on gfn-14 but wishing it was over with!! lol


Totally hear ya on that one! We are really ripping through it though; just a few more months probably. The new WW app might reduce participation a bit, but I'll be on it full force until it is finished :)

For the time I've pretty much abandoned PG stuff; so many other things to sieve and test and find O.O

Well that’s one (a good) way to look at it. Thank you!


:D I try! I've got a prime addiction, so whenever I see a number of interest, I try to find related primes. Got another one too, although it is only a probable prime; proving the primality would be very time consuming (ie. weeks). If you aren't aware of the pop culture significance of the number 1337, take a look here: https://en.wikipedia.org/wiki/Leet and if you are then this is good. Found a PRP with both 484242 and 1337 in it: 484242^1024+1337^1024 is probably prime :)
7) Message boards : Storm message board (Message 144909)
Posted 24 days ago by Kellen
Also! Congrats on the DC! I saw your post on Discord "484,242 digits long. Cool number but a very unprime number of digits", and I must say that I agree that it is quite the composite number of digits, however we can spruce it up a bit ;)

If we multiply 484242 by itself 756 times, then by 3, then subtract 1 we get a prime. Therefore I conclude that 484242 is a component of a very prime number :)

3*484242^756-1 is prime (4298 digits)

Enjoy :)
8) Message boards : Storm message board (Message 144908)
Posted 24 days ago by Kellen
Nice! The 3900XT will make quick work of the challenge tasks :)

I plan to run the Cullen tasks, but only because I already have my badge goal for Woodall. Cullen tasks will have fewer credits per task, but will also take less time to run so you may be able to squeeze a few more in than Woodall tasks through the course of the challenge. I don't think it will matter much though which you run as far as total credits go; pick the one you like the most :)

The only real difference between the two is that you are more likely to find a Woodall prime than a Cullen prime in any given range of k and n values. The chances are near-zero with those anyway though, so it shouldn't make a big difference, lol.
9) Message boards : Storm message board (Message 144906)
Posted 24 days ago by Kellen
Also! Renix! Congrats on your many primes lately!!

Still knocking it out of the park with GFN-14, but I see you on the front page 3 times right now! Lookin pretty good :)
10) Message boards : Storm message board (Message 144905)
Posted 24 days ago by Kellen
Nice! Glad you got it all working :)

As far as finding out FFT size; you can use LLR downloaded from here: http://jpenne.free.fr/ to do so. I am sure that there are other ways, but this is what I use for it for things like this: https://www.primegrid.com/forum_thread.php?id=9339

When you open the program you can select an input from the menu. Make a file called whatever you like (I use "FFTCalc.txt", but it doesn't matter as long as there are no spaces or strange characters) and put the following in it:

ABC $a*$b^$c+$d
1425 2 1660000 1

Then run the LLR software on that file and it will briefly show you the FFT size for the PPSE test that has been yet to be performed of 1425*2^1660000+1 at the beginning of the test. It may also do some trial divisions and tell you that is divisible by some small number. In that case, add 1 to the exponent and try again :)

It is somewhat painstaking to do this to a large set of numbers (that chart took a few hours to generate), but if you just want a basic idea of where the FFT size is on any given project, this works well. For projects like PPSE where a large range of values are being tested, I recommend using the highest k value as it will give the largest FFT size, and that is what I gear my threading towards. So if you want to know FFT size for PPSE, use 9999*2^n+1, instead of 1201*2^n+1.

For challenges, I like to frontload by a bunch and plan for what the FFT size will be by the end of the challenge. I will be running CUL with the assumption that it will be 2048K by challenge end, meaning 2 tasks on 6 cores each for the 3900X

Hopefully this all makes sense. I'm only borderline conscious at the moment so if anything doesn't make sense, just let me know, lol.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2020 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 4.52, 4.48, 4.06
Generated 25 Nov 2020 | 13:02:17 UTC