Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummerslowrise

1)
Message boards :
Fermat Divisor Search :
Fermat Divisor Search
(Message 148683)
Posted 89 days ago by Myrskylyhty
I wonder if after PPS reaches mega levels, it could be possible just to have a PPS project where you select the (range of) k's you want to crunch? No need for separate projects if that would be possible. Of course we would still work from the lowest n up.

2)
Message boards :
Fermat Divisor Search :
Fermat Divisor Search
(Message 148540)
Posted 93 days ago by Myrskylyhty
PS! For some weird reason, k=121 is currently at n = 9.8M, i.e. at higher numbers than we are planning to take PPSDIV to.
That's probably because of the prpnet 121 project http://prpnet.primegrid.com:12001/ . Are k=27 and k=121 ever going to move to BOINC?
A focused PPS project that would run the low k's would be interesting, but would probably require switching the PPS sieve to 9M12M. Another choice is to include the k*2^n1 forms like in 321, but then there would not be chances for DIV primes.
Has there been any discussion whether Primegrid would take a concentrated effort for k*2^n1, or is that left for solo Riesel Prime search at https://www.mersenneforum.org/showthread.php?t=18255?

3)
Message boards :
Number crunching :
Tour de Primes 2021
(Message 147830)
Posted 110 days ago by Myrskylyhty
I posted this in another thread, but I did some quick test runs with my 8700K for which project would be best for searching primes larger than mega primes, and this might interest some folks here.
So if you're looking for a large prime larger than a mega with a CPU, here are your choices (quick test gives a ballpark estimate of "efficiency", with lowest and highest k):
DIV:
Starting Proth prime test of 5*2^8500001+1
Using allcomplex FMA3 FFT length 480K, Pass1=384, Pass2=1280, 6 threads, a = 3
5*2^8500001+1, bit: 60000 / 8500003 [0.70%]. Time per bit: 0.306 ms.
(Total time 43 min)
Starting Proth prime test of 49*2^8500000+1
Using allcomplex FMA3 FFT length 512K, Pass1=512, Pass2=1K, 6 threads, a = 3
49*2^8500000+1, bit: 60000 / 8500005 [0.70%]. Time per bit: 0.324 ms.
(Total time 46 min)
321:
Starting Proth prime test of 3*2^16900001+1
Using allcomplex FMA3 FFT length 960K, Pass1=384, Pass2=2560, 6 threads, a = 5
3*2^16900001+1, bit: 40000 / 16900002 [0.23%]. Time per bit: 0.579 ms.
(Total time 2 h 43 min)
SR5:
Starting N1 prime test of 3622*5^3300000+1
Using allcomplex FMA3 FFT length 640K, Pass1=512, Pass2=1280, 6 threads, a = 3
3622*5^3300000+1, bit: 130000 / 7662373 [1.69%]. Time per bit: 0.425 ms.
(Total time 54 min)
Starting N1 prime test of 152588*5^3300001+1
Using allcomplex FMA3 FFT length 800K, Pass1=640, Pass2=1280, 6 threads, a = 3
152588*5^3300001+1, bit: 40000 / 7662380 [0.52%]. Time per bit: 0.541 ms.
(Total time 1 h 9 min)
TRP:
Starting Lucas Lehmer Riesel prime test of 2293*2^115000011
Using FMA3 FFT length 864K, Pass1=384, Pass2=2304, 6 threads
V1 = 4 ; Computing U0...done.
2293*2^115000011, iteration : 30000 / 11500001 [0.26%]. Time per iteration : 0.711 ms.
(Total time 1 h 53 min)
Starting Lucas Lehmer Riesel prime test of 494743*2^115000011
Using zeropadded FMA3 FFT length 1200K, Pass1=320, Pass2=3840, 6 threads
V1 = 4 ; Computing U0...done.
494743*2^115000011, iteration : 80000 / 11500001 [0.69%]. Time per iteration : 0.779 ms.
(Total time 2 h 29 min)
For comparison, here is results with PPSMega and PPSMega with DIVsize n:
Starting Proth prime test of 1201*2^3350000+1
Using allcomplex FMA3 FFT length 256K, Pass1=128, Pass2=2K, 6 threads, a = 3
1201*2^3350000+1, bit: 70000 / 3350010 [2.08%]. Time per bit: 0.172 ms.
(Total time ~9 min)
Starting Proth prime test of 9999*2^3350001+1
Using allcomplex FMA3 FFT length 256K, Pass1=128, Pass2=2K, 6 threads, a = 13
9999*2^3350001+1, bit: 70000 / 3350014 [2.08%]. Time per bit: 0.170 ms.
(Total time ~9 min)
Starting Proth prime test of 1201*2^8500000+1
Using allcomplex FMA3 FFT length 640K, Pass1=512, Pass2=1280, 6 threads, a = 3
1201*2^8500000+1, bit: 50000 / 8500010 [0.58%]. Time per bit: 0.408 ms.
(Total time 58 min)
Starting Proth prime test of 9999*2^8500001+1
Using allcomplex FMA3 FFT length 768K, Pass1=384, Pass2=2K, 6 threads, a = 7
9999*2^8500001+1, bit: 110000 / 8500014 [1.29%]. Time per bit: 0.460 ms.
(Total time 1 h 5 min)
So, DIV is a good choice if you're aiming for anything over Mega (SR5 can be also good, it has lower digits = higher chance of prime per task).

4)
Message boards :
Fermat Divisor Search :
Fermat Divisor Search
(Message 147827)
Posted 110 days ago by Myrskylyhty
Less if people crunch it more during TdP, since the low k's offer higher "efficiency" for large primes.
Unless I’m missing something, that’s incorrect.
I’m pretty sure that the value of k has no effect on the likelihood of the number being prime. A low k increases the chance of a prime being a Fermat divisor, but that only applies if the number is prime.
PPSDIV numbers are too large, at 2.5 million digits, to be a good choice if you’re just trying to get a prime or a mega prime. It might be a good choice if you’re going for the largest prime jersey.
Yep I meant the fact that lower k are more efficient to compute compared to higher k with same n. Therefore its nicely efficient if you want to reach for the largest prime jersey, but dont want take your odds with 321. PPS mega is of course better if you're looking just for a mega :)
So if you're looking for a large prime larger than a mega with a CPU, here are your choices (quick test gives a ballpark estimate of "efficiency", with lowest and highest k):
DIV:
Starting Proth prime test of 5*2^8500001+1
Using allcomplex FMA3 FFT length 480K, Pass1=384, Pass2=1280, 6 threads, a = 3
5*2^8500001+1, bit: 60000 / 8500003 [0.70%]. Time per bit: 0.306 ms.
(Total time 43 min)
Starting Proth prime test of 49*2^8500000+1
Using allcomplex FMA3 FFT length 512K, Pass1=512, Pass2=1K, 6 threads, a = 3
49*2^8500000+1, bit: 60000 / 8500005 [0.70%]. Time per bit: 0.324 ms.
(Total time 46 min)
321:
Starting Proth prime test of 3*2^16900001+1
Using allcomplex FMA3 FFT length 960K, Pass1=384, Pass2=2560, 6 threads, a = 5
3*2^16900001+1, bit: 40000 / 16900002 [0.23%]. Time per bit: 0.579 ms.
(Total time 2 h 43 min)
SR5:
Starting N1 prime test of 3622*5^3300000+1
Using allcomplex FMA3 FFT length 640K, Pass1=512, Pass2=1280, 6 threads, a = 3
3622*5^3300000+1, bit: 130000 / 7662373 [1.69%]. Time per bit: 0.425 ms.
(Total time 54 min)
Starting N1 prime test of 152588*5^3300001+1
Using allcomplex FMA3 FFT length 800K, Pass1=640, Pass2=1280, 6 threads, a = 3
152588*5^3300001+1, bit: 40000 / 7662380 [0.52%]. Time per bit: 0.541 ms.
(Total time 1 h 9 min)
TRP:
Starting Lucas Lehmer Riesel prime test of 2293*2^115000011
Using FMA3 FFT length 864K, Pass1=384, Pass2=2304, 6 threads
V1 = 4 ; Computing U0...done.
2293*2^115000011, iteration : 30000 / 11500001 [0.26%]. Time per iteration : 0.711 ms.
(Total time 1 h 53 min)
Starting Lucas Lehmer Riesel prime test of 494743*2^115000011
Using zeropadded FMA3 FFT length 1200K, Pass1=320, Pass2=3840, 6 threads
V1 = 4 ; Computing U0...done.
494743*2^115000011, iteration : 80000 / 11500001 [0.69%]. Time per iteration : 0.779 ms.
(Total time 2 h 29 min)
For comparison, here is results with PPSMega and PPSMega with DIVsize n:
Starting Proth prime test of 1201*2^3350000+1
Using allcomplex FMA3 FFT length 256K, Pass1=128, Pass2=2K, 6 threads, a = 3
1201*2^3350000+1, bit: 70000 / 3350010 [2.08%]. Time per bit: 0.172 ms.
(Total time ~9 min)
Starting Proth prime test of 9999*2^3350001+1
Using allcomplex FMA3 FFT length 256K, Pass1=128, Pass2=2K, 6 threads, a = 13
9999*2^3350001+1, bit: 70000 / 3350014 [2.08%]. Time per bit: 0.170 ms.
(Total time ~9 min)
Starting Proth prime test of 1201*2^8500000+1
Using allcomplex FMA3 FFT length 640K, Pass1=512, Pass2=1280, 6 threads, a = 3
1201*2^8500000+1, bit: 50000 / 8500010 [0.58%]. Time per bit: 0.408 ms.
(Total time 58 min)
Starting Proth prime test of 9999*2^8500001+1
Using allcomplex FMA3 FFT length 768K, Pass1=384, Pass2=2K, 6 threads, a = 7
9999*2^8500001+1, bit: 110000 / 8500014 [1.29%]. Time per bit: 0.460 ms.
(Total time 1 h 5 min)
So, DIV is a good choice if you're aiming for anything over Mega (SR5 can be also good, it has lower n = higher chance of prime per task).

5)
Message boards :
Fermat Divisor Search :
Fermat Divisor Search
(Message 147819)
Posted 111 days ago by Myrskylyhty
With 342589 tasks left with current speed of 6390 tasks/per day, we have 53.6 days of work left.
Less if people crunch it more during TdP, since the low k's offer higher "efficiency" for large primes.
So, we might easily finish DIV during TdP!

6)
Message boards :
Fermat Divisor Search :
Fermat Divisor Search
(Message 147676)
Posted 116 days ago by Myrskylyhty
Less than 400 000 tasks left! I think I'll run this until TdP. Maybe even during TdP?

7)
Message boards :
321 Prime Search :
321 Sieve is being SUSPENDED
(Message 143721)
Posted 230 days ago by Myrskylyhty
Maybe when us youngsters are at a retirement home, we'll start up 321 sieve for 50M100M.

8)
Message boards :
Number crunching :
Badges III
(Message 139090)
Posted 418 days ago by Myrskylyhty
Finally! 321 is a blast!

9)
Message boards :
Proth Prime Search :
PPSMEGA: Smaller FFT longer crunch time ?
(Message 138158)
Posted 447 days ago by Myrskylyhty
With my 9900KS, the 240K FFT length PPS Mega tasks take over 2x longer than the 256K FFT length tasks. Even DIV tasks are faster than the 240K FFT length PPS Mega.
What's even more weird: trying to run any current length PPS Mega task using the independant LLR app in command line, every single task runs with 256K FFT length.
Could something cause the Boinc LLRapp to occasionally run using a "wrong" FFT length?

10)
Message boards :
Number crunching :
Tour de Primes 2020
(Message 137827)
Posted 457 days ago by Myrskylyhty
Just a note to people: Even if your 1st % goes up by multithreading, your 1st task rate might not necessarily go up.
For example, a 9900KS will go through 30 % more tasks if singlecoring PPSE rather than twocoring. That means if I wanted to use two cores per tasks, my 1st percentage would have to go up over 30 % to offset the difference.
In addition there is the fact that the first option is processing 30 % more candidates, thus advancing our work as a Primegrid collective much more! So if you're breaking even, please use the more productive option.
I think people are a bit blinded by just the 1st percentage rather than looking at the 1st tasks per day. It is a bit of a silly situation if everyone just gives up 30 % of their computing power while the effective prime finder chance stays the same.
Play it smart everyone! :)

Next 10 posts
