[QUOTE=Mystwalker]That was until v24. Now, there's no difference in speed...[/QUOTE]

Fine! :smile:

Luigi

[QUOTE=Mystwalker]That was until v24. Now, there's no difference in speed...[/QUOTE]

No difference is not quite right, although now that version 24 includes SSE2 support for 2^n+1 numbers, the difference is much smaller than it used to be for Pentium-4's and other computers. Perhaps George should change the note on the Fermat number factoring status page, as for most numbers, it is now more efficient to work on 2^n+1 forms. However, I am still running curves on M2097152 which, on both my Athlon and P4, is more efficient than running separate curves on F17, F18, F19, and F20. I think that the reason for this is that the FFT size is 112K, wheras the curves on F17 through F20 use FFT sizes of 8K, 16K, 32K, and 64K. On the other hand, F12, F13, F14, F15, and F16 use only 3/4 of the size they did previously, so for example, F16 uses 3K instead of 4K. So the small Fermat numbers are much more efficient with version 24, the larger ones only moderately so, except with Pentium 4's. On the other hand, for large Fermat numbers, F20 and higher, the amount of memory allocated during stage 2 makes a big difference, so it may actually be more efficient to run separate curves in some cases. My advice is, run separate curves for small numbers and for large numbers, benchmark to find out what is most efficient.

Is it normal?
 

I started work on M131072 with B1=1000000, B2=default, just for curiosity. After a few hours I noticed, that PRIME95 is working on curve #7, and...

[Jul 23 23:28] At prime 34903909. Time thusfar 2066.090 sec.
etc.

Why is it working on such big prime, exceeding both B1 and B2?
I'm running newest Prime95, on Athlon 2000+...

Probably I am missing something, so please explain it shortly.

And one more thing: where does it report any factor found? I also have noticed that results.txt isnt't touched by the program while running these curves...

Washuu

PS. It's already at prime 40653281. :whistle:

[QUOTE=Washuu]Why is it working on such big prime, exceeding both B1 and B2?[/QUOTE]
Default B2 = 100B1. So it is well within that limit.

Factors found will be shown both in the screen and recorded in results.txt.

[QUOTE=Washuu]I started work on M131072 with B1=1000000, B2=default.
[/QUOTE]

Note that M131072 = 2^131072 -1 is the product of all the Fermat numbers F0 through F16 (=P65536). The status page shows 1800 curves already run for these numbers at B1=1000000. If you want to run curves on F17 = P131072, you need to run on 2^131072 +1 instead. (I am currently running 100 curves on this number, so another 132 curves will finish it up at this level.)

[QUOTE=Washuu]
And one more thing: where does it report any factor found? I also have noticed that results.txt isnt't touched by the program while running these curves...
[/QUOTE]

When you finish the 100 curves, or whatever number you specified, Prime95 will write a line in results.txt saying that you ran 100 curves, whether or not you found a factor.

If you have some memory to spare, you should use GMP-ECM version 6 to do stage 2 for smaller Fermat numbers. The Schönhage-Strassen multiplication speeds that up quite a bit. You can link in George's GWNUM library as well, but this is a bit tricky for 6.0. In 6.1, it will be easier and the GWNUM stage 1 will be incorporated as well.

Alex

I did 130 curves of 2^32768-1 with B1=11,000,000 and B2=110,000,000. Which mail adress do I report this to? [email]woltman@alum.mit.edu[/email] ?


Doing curves at B1=43,000,000 now, but might go back and do more at B1=11,000,000 again.

[QUOTE=ATH]I did 130 curves of 2^32768-1 with B1=11,000,000 and B2=110,000,000. Which mail adress do I report this to? [email]woltman@alum.mit.edu[/email] ?


Doing curves at B1=43,000,000 now, but might go back and do more at B1=11,000,000 again.[/QUOTE]

Yessir :smile:

I just finished 120 curves at the same number... :wink:

Luigi

Two things - the new version 24 of Prime95 uses SSE2 instructions on P16384, so the old note on the ecmf.html webpage about running on M32768 with a Pentium-4 being faster is probably not true any more. You could benchmark a curve on each to see which is faster.

Also, have either of you tried using GMP-ECM on stage 2 ? I have a computer with 512Mb of memory, so I'll try it soon and see how it works.

Configuration for Linux
 

Hello Philmoore,
[QUOTE=philmoore]I am assuming that you are running on computers running some sort of Windows operating system, but if not, feel free to ask questions here. [/QUOTE] 1) Would you mind providing the explainations for Linux O.S. ? or pointing me to a web-page.
2) Also, is this thread the place for reserving parameters for F14 ? If not, where is it ?
3) About page [URL=http://www.mersenne.org/ecmf.htm]Factoring Fermat Numbers[/URL] , I would suggest to add a column on the left with the Fermat names (F12, F13, F14, ...) added to the column with N (2^N+1). (It took me some time understanding that 16384 was meaning F14 ...).

I'd like to contribute to the search of a factor of F14.

Thanks,
Tony

I don't know much about Linux, so I'll leave your first question to someone else. I think that you have to edit the worktodo file to do ECM with mprime, but I am not sure what is involved if you want to run the second stage with GMP-ECM.

The most efficient factoring methods at this point are P-1 and ECM. P-1 has already been run to a high range, so ECM is probably the best method for now. Because ECM, as run by either the programs Prime95 (or the linux version mprime) or GMP-ECM, runs on randomly generated curves, several people can run ECM simultaneously, as the chances that they are running on the same curve are extremely small. So there is no need to reserve parameters. Simplest thing if you are running mprime is to put a line such as
ECM=16384,11000000,0,100,0,0,1,0
into your worktodo file. (This command will run 100 curves on F14 to a stage 1 bound of 11000000 and a stage 2 bound of 100*(stage 1 bound).)

I'll pass along your suggestion about identifying the Fermat numbers explicitly to George.