Hi,

I ran a short test on FPS (range 8844-8845G) with the CPU and GPU app, but my GPU returned more results:

fsievecl64.exe:
---------------
8844124870297 | 23288!-1
8844301414811 | 490811!+1
8844346130771 | 413493!+1
8844391565431 | 453319!-1
8844512918177 | 379163!-1
8844516749147 | 695243!+1
8844635693029 | 313352!+1
8844792669553 | 766503!+1
8844821508923 | 579427!-1
8844830308543 | 719502!+1
8844963952049 | 692143!+1

fpsieve-x86_64-linux:
---------------------
8844346130771 | 413493!+1
8844516749147 | 695243!+1
8844635693029 | 313352!+1
8844792669553 | 766503!+1
8844821508923 | 579427!-1
8844963952049 | 692143!+1

I'm not into the topic, so my question is: Is it ok that the CPU app "potentially" removes less candidates (and sometimes even 0) than the OpenCL app?

I ask because my CPU took about 3 hours for 1G whereas my GPU crunched a little under an hour for the same range.

The difference is due to the sieve file used as input for the CPU sieving program. The CPU program finds all the same factors, but suppresses output for anything that's already been removed from the sieve. If you look at the factorial statistics page, you'll see that some entries show a large difference between the factors found and the candidates removed. Those were run on GPU. The ones where the factors and removals are identical or very close were run on CPU. No matter which you run, the candidates removed number will be identical.

If you want to see this in action, run the CPU range again but leave out the -i fsieve_7T.txt part from the command line. You'll get the exact same output as the GPU program.

Edit: Actually I believe the CPU program doesn't even bother testing anything that's not in the sieve file and so it runs faster with a sieve file as input.

Thank you very much for making this clear because I wasn't sure about the behaviour. I could have done more testing, yes... but... ^^