mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2020-08-26, 17:11   #1
bur
 
Aug 2020

2016 Posts
Default For which types of primes is GPU primality test software available?

Which types of primes can be (sort of efficiently) tested for primality using GPU software? Or rather, for which types is software available?


I know of generalized Fermat and Mersenne. Also arithmetic progressions. Is there software for other primes, e.g. Proth primes? Or their -1 counterpart (is there a name for those?)?



The reason I'm asking is that I'd like to do some manual prime hunting (currently doing Primegrid) but I don't want to move CPUs away from PG. So I thought of buying a card like a 1650 and use it to play around. But for primes that are already covered by either PG or GIMPS it doesn't really make sense to work individually.


Proth primes are also searched by PG, but it's easy to choose a k not within their range.
bur is offline   Reply With Quote
Old 2020-08-26, 18:50   #2
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

446610 Posts
Default

Quote:
Originally Posted by bur View Post
Or their -1 counterpart (is there a name for those?)?
The -1 side is called Riesel.

llrCUDA can test Riesel form, but the software is best described as "alpha".
Have a peek at https://mersenneforum.org/showthread.php?t=14608 for some details on that. If you're feeling adventurous, you could perhaps get it working; I've done so, but with less than impressive speed.

Now that GPUowl is doing prp tests, I wonder if it could easily be adapted to run prp tests on riesel inputs, or proth for that matter. Getting a "probable prime" result from the GPU could then be proven with the usual CPU software packages available.

Last fiddled with by VBCurtis on 2020-08-26 at 18:51
VBCurtis is online now   Reply With Quote
Old 2020-08-27, 08:00   #3
henryzz
Just call me Henry
 
henryzz's Avatar
 
"David"
Sep 2007
Cambridge (GMT/BST)

5,743 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
The -1 side is called Riesel.

llrCUDA can test Riesel form, but the software is best described as "alpha".
Have a peek at https://mersenneforum.org/showthread.php?t=14608 for some details on that. If you're feeling adventurous, you could perhaps get it working; I've done so, but with less than impressive speed.

Now that GPUowl is doing prp tests, I wonder if it could easily be adapted to run prp tests on riesel inputs, or proth for that matter. Getting a "probable prime" result from the GPU could then be proven with the usual CPU software packages available.
I think llr testing is on the long todo list for GPUowl but is one of those things that never seems to reach the top. How hard is it to add this?
henryzz is offline   Reply With Quote
Old 2020-08-27, 17:25   #4
bur
 
Aug 2020

25 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
The -1 side is called Riesel.
I thought Riesel numbers are those k in k*2^n-1 that never produce primes. Or is the whole construct also called Riesel number?


Quote:
Now that GPUowl is doing prp tests, I wonder if it could easily be adapted to run prp tests on riesel inputs, or proth for that matter. Getting a "probable prime" result from the GPU could then be proven with the usual CPU software packages available.
That sounds like a very interesting idea, Proth sieving can be done on GPU with high efficiency, so if GPUowl could indeed PRP the candidates, then only a few remaining numbers would have to be LLR checked with CPU.


Quote:
Originally Posted by henryzz
I think llr testing is on the long todo list for GPUowl but is one of those things that never seems to reach the top.
According to rieselprime.de, GPUowl supports LL?


What are the chances PRP checking Proth/Riesel will be implemented? Is there active development on GPUowl? How can the coder(s) be approached?



If all else fails, I might just do the sieving using GPU and then LLR the candidates with CPU.
bur is offline   Reply With Quote
Old 2020-08-27, 18:07   #5
Viliam Furik
 
Jul 2018
Martin, Slovakia

3708 Posts
Default

Quote:
Originally Posted by bur View Post
I thought Riesel numbers are those k in k*2^n-1 that never produce primes. Or is the whole construct also called Riesel number?
Yes, Riesel number is k in k*2^n-1, but when the number in this form is prime, it is called Riesel prime. Also, I think that the numbers themselves are called Riesel numbers (depends on the context)

Quote:
Originally Posted by bur View Post
According to rieselprime.de, GPUowl supports LL?
Yes, it does support LL - Lucas-Lehmer test. But what you need for the Proth and Riesel primes is LLR - Lucas-Lehmer-Riesel test, which is almost the same except you need to determine the starting value, which is 4 in LL, but other numbers in LLR.

Quote:
Originally Posted by bur View Post
What are the chances PRP checking Proth/Riesel will be implemented? Is there active development on GPUowl? How can the coder(s) be approached?
Coder is Mihai Preda (forum username: preda)
Viliam Furik is offline   Reply With Quote
Old 2020-08-27, 20:42   #6
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

37·127 Posts
Default

Welcome to the gpu side.
There's a considerable wish list in the queue for gpuowl. See https://www.mersenneforum.org/showpo...35&postcount=6
Plus more, that did not make that poll.
Lots of info on gpuowl at its reference thread https://www.mersenneforum.org/showthread.php?t=23391
A GTX1650 is excellent at trial factoring (which does not currently involve gpuowl). See mfaktc for that.
A 5700XT is decent for gpuowl. Or RadeonVII is excellent, if you can find one at an acceptable price.

Last fiddled with by kriesel on 2020-08-27 at 21:13
kriesel is offline   Reply With Quote
Old 2020-08-28, 06:20   #7
bur
 
Aug 2020

1000002 Posts
Default

Ok, so PRP for Proth didn't make the list... I read in a post from 2016 or so that generally GPUs aren't that great for LLR. And there are many tasks such as TF of Mersenne or PRP of GFN where they excel so apparently it doesn't make sense to waste them on LLR.


But will GPUs generally be able to efficiently do a PRP test on Proth or Riesel numbers? It would be the same approach PG takes for GFN. GPUs are used for doing the PRP test and if positive a CPU checks with LLR. According to PG stats testing a GFN-18 takes less than one hour on GPU and more than 30 hours on CPU. If similar results could be obtained for Proth/Riesel it would speed things up considerably.


I have no idea though if it's just a specific property of GFNs that makes GPUs that efficient.
bur is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Fastest software for Mersenne primality test? JonathanM Information & Answers 25 2020-06-16 02:47
I found the primality test, there seems to be no composite numbers that pass the test sweety439 sweety439 7 2020-02-11 19:49
Modifying the Lucas Lehmer Primality Test into a fast test of nothing Trilo Miscellaneous Math 25 2018-03-11 23:20
Stress Test Error Types? TheMawn Information & Answers 0 2014-07-22 00:27
Primality-testing program with multiple types of moduli (PFGW-related) Unregistered Information & Answers 4 2006-10-04 22:38

All times are UTC. The time now is 07:15.

Tue Nov 24 07:15:51 UTC 2020 up 75 days, 4:26, 4 users, load averages: 1.48, 1.44, 1.52

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.