Benchmark of current tools
 

Hello,

Just as a piece of information or comparison, I've made a benchmark including:
-1) 3 tools running on my laptop
-2) 2 curves from this [URL="http://www.mersenneforum.org/showpost.php?p=423225&postcount=9"]thread[/URL] (as a marker of reference)
My computer is an ACER-V3 771G with an Intel Core i5 3230M, 8GiB RAM, around 20.8 GFLOPS @ 2.6 GHz & it can reach up to 25.6 GFLOPS @ 3.2 GHz.
The two last curves, in purple and orange, are an adaptation of regressions gived in the post I mentioned. Those are calculated according to a digits-length of the integer, my graph is based on a bits-length. That's why there are actually 2 purple lines, they come from the nearest approximation we can make according to conversion from base 10 to base 2.

I've set the maximum possible number of cores used during the factorization when it was possible, otherwise I suppose the settings have been given automatically by default by the tools.

[URL="http://www.hostingpics.net/viewer.php?id=983143stats.png"]http://hpics.li/a019fea[/URL]

Clearly, this is not supposed to be any kind of reference, I only made it for test purpose and to get a precise approximation of calculus time for my laptop. But I share it just for fun.

New version
 

New Version:
With some improvements:
-Accuracy of regressions
-More measures
-Old decimal-based curves removed

However we've to notice that CADO-NFS' curve isn't relevant at all. An error on my current compiled one prevented me from doing new tests. I keeped it anyway.

We can also see that YAFU is indeed and obviously the fastest program ever.

I've projects: do the same tests on sagemath and maybe others, and with mersenne numbers (except prime ones of course).

All tested numbers are generated by typing rsa(b) in YAFU but are not necessarily the same numbers for each tool, although the bitlength of the RSA-number are the same.

[URL]http://hpics.li/0796437[/URL]