## Other

drummers-lowrise

Message boards : Generalized Fermat Prime Search : GFN CPU limits

Author Message
Yves Gallot
Volunteer developer
Project scientist

Joined: 19 Aug 12
Posts: 820
ID: 164101
Credit: 305,989,513
RAC: 1,728

Message 158880 - Posted: 25 Dec 2022 | 11:10:25 UTC

Santa Claus brought a new algorithm!

A quick implementation found that 1,010,036,096216 + 1 is prime, error = 0.140625.
The current limit of GFN-16 on CPU is about 180M. With the new method, it will be greater than 1G.
The computation time of the final optimized version should be below +20% of the current version.
GFN-15 will be able to run on CPU: 323,997,566215 + 1 is prime, error = 0.09375.

Merry Christmas!

Dave

Joined: 13 Feb 12
Posts: 3209
ID: 130544
Credit: 2,288,183,124
RAC: 754,139

Message 158881 - Posted: 25 Dec 2022 | 11:18:07 UTC - in response to Message 158880.

Wowsers, the doorstops are back in the game!

Crun-chi
Volunteer tester

Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 54,956

Message 158882 - Posted: 25 Dec 2022 | 11:31:55 UTC - in response to Message 158880.

Santa Claus brought a new algorithm!

A quick implementation found that 1,010,036,096216[/sup] + 1 is prime, error = 0.140625.
The current limit of GFN-16 on CPU is about 180M. With the new method, it will be greater than 1G.
The computation time of the final optimized version should be below +20% of the current version.
GFN-15 will be able to run on CPU: 323,997,5662[sup]15 + 1 is prime, error = 0.09375.

Merry Christmas!

Somehow this prime is known for me :)
But I am proud to help :)
Great thing!

____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie!

Michael Millerick
Volunteer tester

Joined: 4 Feb 09
Posts: 938
ID: 35074
Credit: 915,078,072
RAC: 1,434,967

Message 158883 - Posted: 25 Dec 2022 | 11:54:01 UTC

This is excellent news! Hopefully it hits in time to keep GFN16 on CPU a viable way to compete in TdP this year.
____________

Michael Millerick
Volunteer tester

Joined: 4 Feb 09
Posts: 938
ID: 35074
Credit: 915,078,072
RAC: 1,434,967

Message 159027 - Posted: 2 Jan 2023 | 2:49:44 UTC

Is the new application only going to be installed on the server when we get closer to the limit of the current application on GFN16?
____________

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 940
ID: 3110
Credit: 261,913,874
RAC: 8,863

Message 159198 - Posted: 8 Jan 2023 | 18:17:03 UTC

I hope it gets installed soon. We're over 180M on GFN-16 now and I'm starting to see CPU errors. This one errored on two very different CPUs before a GPU picked it up.

mikey

Joined: 17 Mar 09
Posts: 1788
ID: 37043
Credit: 793,513,261
RAC: 1,209,755

Message 159210 - Posted: 9 Jan 2023 | 0:26:17 UTC - in response to Message 158880.

Santa Claus brought a new algorithm!

A quick implementation found that 1,010,036,096216 + 1 is prime, error = 0.140625.
The current limit of GFN-16 on CPU is about 180M. With the new method, it will be greater than 1G.
The computation time of the final optimized version should be below +20% of the current version.
GFN-15 will be able to run on CPU: 323,997,566215 + 1 is prime, error = 0.09375.

Merry Christmas!

So as this gets implemented does that mean some the 'ending soon' dates will be pushed as well, that's in addition to the GRN-15 and 16 tasks?

Yves Gallot
Volunteer developer
Project scientist

Joined: 19 Aug 12
Posts: 820
ID: 164101
Credit: 305,989,513
RAC: 1,728

Message 159213 - Posted: 9 Jan 2023 | 9:30:18 UTC - in response to Message 159210.

So as this gets implemented does that mean some the 'ending soon' dates will be pushed as well, that's in addition to the GFN-15 and 16 tasks?

If there are some 'ending soon' dates somewhere, you can forget them. Now the limit is two billion (2G) on both CPU and GPU for all exponents.

In the range [2; 2G] different algorithms can be implemented: when b is small, some are faster but they can be applied to a limited range.
The end of a range is not a clear boundary (because of floating-point numbers, it is fuzzy). The probability of an error is increasing slowly with b. For this reason, the limits are not defined in the code but genefer is generated for the current leading edges of PrimeGrid.

genefer 4.01 is not able to test GFN-15 on CPU and is starting to generate errors for GFN-16. genefer 4.02 extends the bound to b ~ 2G for GFN-15 and GFN-16. This version is already in production for GFN-17 (see https://www.primegrid.com/apps.php), I think that it will soon be rolled out for other exponents.

Yves Gallot
Volunteer developer
Project scientist

Joined: 19 Aug 12
Posts: 820
ID: 164101
Credit: 305,989,513
RAC: 1,728

Message 159245 - Posted: 10 Jan 2023 | 20:06:59 UTC - in response to Message 159198.