David Cleaver wrote:
> Is it better to tackle big numbers like this with Prime95 or ecm-5.0.
I did some quick testing on a P4
For stage 1, Prime95 is MUCH faster than GMP-ECM 5
I tried running a few curves with B1=1e6 on M(971) which has no known
factors (293 digits), and on M(959) which has a few known factors (254
digits, i didn't spent time looking for 1 or a few more known factors which
make it a C213).
M(971) Prime95 ~28.5
GMP-ECM 5 ~102
M(959) Prime95 ~28.3
GMP-ECM 5 ~113
Even though M(959) has known factors, GMP-ECM seems faster running on the
For stage 2 things are a bit different. Prime95 uses a default B2=100*B1
while GMP-ECM uses a much larger B2 (in my case for B1=1M, B2 was ~839M)
M(971) Prime95 ~15.5
GMP-ECM 5 ~16.9
M(959) Prime95 ~15.5
GMP-ECM 5 ~14
With the default B2, GMP-ECM needs 75 resp. 64 seconds for 1 curve but has a
higher chance of finding a factor.
So only when a numbers has known factors, B2 might be faster with GMP-ECM
(it is possible to run B1 with prime95 and B2 with GMP-ECM).
On a P3 or Athlon things might be very different since they don't use the
SSE2 optimizations in Prime95
> And, how does it count as an ecm record? Does Prime95 run ecm curves on
> numbers, and if so, how do you know what size factors it is looking
You can set the bounds youreself in Prime95