## Other

drummers-lowrise

Message boards : Fermat Divisor Search : Computation times vary strongly

 Post to thread Subscribe SortOldest firstNewest firstHighest rated posts first
Author Message
Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 395
ID: 1241833
Credit: 149,627,705
RAC: 947,715

Message 144531 - Posted: 21 Oct 2020 | 17:50:42 UTC

During the challenge I noticed that the computation times per task vary a lot. All computers are set single-threaded and one task per core (50% CPU on Host 1 to prevent HT).

Avg. / Min. / Max.
Host 1 (14 tasks / 24 h)
24,029 / 13,047 / 31,047

Host 2 (14 tasks / 24 h)
21,670 / 19,399 / 25,825

Host 3 (8 tasks / 24 h)
20,807 / 20,260 / 21,318

Host 1 is running GFN 17 mega simultaneously, but host 2 is more or less all the time idle and Host 3 which has the most constant values is actually my desktop computer. Does the time depend that strongly on k?

Furthermore, Host 1 is an i7-4790K, Host 2 is an i5-4590 and Host 3 is an i3-2120. I would have expected the i3 to be much slower, but it isn't.
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Scott Brown
Volunteer moderator
Volunteer tester
Project scientist

Joined: 17 Oct 05
Posts: 2254
ID: 1178
Credit: 9,992,663,109
RAC: 10,765,414

Message 144534 - Posted: 21 Oct 2020 | 17:56:19 UTC - in response to Message 144531.

There are three different fft sizes currently being run on the DIV project: 384k, 400k, 480k. These will have longer run times, respectively.

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 395
ID: 1241833
Credit: 149,627,705
RAC: 947,715

Message 144556 - Posted: 22 Oct 2020 | 6:53:36 UTC - in response to Message 144534.
Last modified: 22 Oct 2020 | 6:53:53 UTC

Thanks, you're right, that was the cause. Surprisingly (to me) it's not the difference in n, which varied from 6.78 to 6.83, but the difference in k that causes FFT size to grow.

So large k will take much longer to compute while having marginal impact on the overall size of the prime?
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

JeppeSN

Joined: 5 Apr 14
Posts: 1497
ID: 306875
Credit: 33,833,412
RAC: 40,756

Message 144558 - Posted: 22 Oct 2020 | 7:49:22 UTC - in response to Message 144556.

So large k will take much longer to compute while having marginal impact on the overall size of the prime?

Correct. Additionally, primes with larger k have smaller chances to be divisors of Fermat numbers and generalized Fermat numbers. /JeppeSN

Bur
Volunteer tester

Joined: 25 Feb 20
Posts: 395
ID: 1241833
Credit: 149,627,705
RAC: 947,715

Message 144652 - Posted: 23 Oct 2020 | 16:15:24 UTC - in response to Message 144558.

I just realized that's exactly what the table posted by Kellen tells us.
____________
Primes: 1281979 & 12+8+1979 & 1+2+8+1+9+7+9 & 1^2+2^2+8^2+1^2+9^2+7^2+9^2 & 12*8+19*79 & 12^8-1979 & 1281979 + 4 (cousin prime)

Message boards : Fermat Divisor Search : Computation times vary strongly