mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   FermatSearch (https://www.mersenneforum.org/forumdisplay.php?f=133)
-   -   Feromant performance (https://www.mersenneforum.org/showthread.php?t=21800)

firejuggler 2016-12-03 18:52

Feromant performance
 
So I did some test with Feromant. On a gtx 750 Ti, the CUDA version seem to go 10 time faster thaan my CPU (i5-2400k)

F CUDA normal
41 286,375,579 28,615,901
81 137,500,927 14,298,381
161 80,650,011 7,125,026
321 36,940,695 2,388,966
479 14,139,185 705,062

F479 is the highest i could use the cuda version without crashing.

ET_ 2016-12-03 19:01

[QUOTE=firejuggler;448313]So I did some test with Feromant. On a gtx 750 Ti, the CUDA version seem to go 10 time faster thaan my CPU (i5-2400k)

F CUDA normal
41 286,375,579 28,615,901
81 137,500,927 14,298,381
160 80,650,011 7,125,026[/QUOTE]

Consider that Feromant_CUDA uses one CPU core for the sieving.

feromant 2016-12-04 07:05

1 Attachment(s)
Hi All!
I have the results of some calculations on different equipment. I think they will be useful.

The program has not been tested for generation Pascal GPU, so I would be interested in information about timings and whether program test run (from a file _worktodo_factors).

Yes, the program fails when the speed is below 15M/sec

ET_ 2016-12-04 12:16

[QUOTE=feromant;448339]Hi All!
I have the results of some calculations on different equipment. I think they will be useful.

The program has not been tested for generation Pascal GPU, so I would be interested in information about timings and whether program test run (from a file _worktodo_factors).

Yes, the program fails when the speed is below 15M/sec[/QUOTE]

Welcome aboard Roman!

Luigi

Batalov 2016-12-05 04:06

[QUOTE=feromant;448339]Hi All!
I have the results of some calculations on different equipment. I think they will be useful.

The program has not been tested for generation Pascal GPU, so I would be interested in information about timings and whether program test run (from a file _worktodo_factors).

Yes, the program fails when the speed is below 15M/sec[/QUOTE]
Can you build a linux CUDA binary (or let someone trusted build it and have it sent back to you, so then you could distribute the binary)? I'd be happy to help.

feromant 2016-12-05 10:40

[QUOTE=Batalov;448412]Can you build a linux CUDA binary (or let someone trusted build it and have it sent back to you, so then you could distribute the binary)? I'd be happy to help.[/QUOTE]

I plan on posting the source code for public use later.


All times are UTC. The time now is 15:49.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.