How much Trial Factoring to do?

[odin]

Hello All,

I've been assigned some exponents in the 50M range to be LL tested. They were Trial factored up to 2^68 and P-1 factored and not factors found.

Would it make sense to manually assign the exponent to be Trial factored up to 2^70 just to make sure it does not have any low factors and then LL test it if it does not?

It takes about 18 hours to Trial factor one increment like from 2^68 to 2^69 so 36 hours to go from 2^68 to 2^70 and it takes 25 days to LL test.

Please tell me if this would be a waste of time.

Thanks!

[Mini-Geek]

If you use the default configuration, (e.g. you use Test= and don't modify the TF'd depth, etc.) Prime95 will automatically choose what should optimally be done. According to http://mersenne-aries.sili.net/facto...tton=Calculate, for an exponent size 50M, Prime95 will factor to 2^69.

[markr]

Hi Odin,

Assignments in the 50M range should be trial factored to 2^69 before the LL test, unless the bounds have changed since version 24.12. George worked out that's optimum in terms of work done overall.

The way it's supposed to happen is: TF to 2^68, P-1, TF to 2^69, LL. I thought the client (mprime or prime95) automatically did any missing factoring steps before the LL, so you shouldn't need to do anything manually. Has your LL test started?

(Mini-Geek beat me to it!)

[lavalamp]

Originally Posted by odin

It takes about 18 hours to Trial factor one increment like from 2^68 to 2^69 so 36 hours to go from 2^68 to 2^70

If it takes you 18 hours to do 68 -> 69, then it will take you 36 hours to do 69 -> 70, for a total of 54 hours.

Each bit increase represents a DOUBLING of the work done in the previous level.

To put it another way, it would take the same amount of CPU time to trial factor from scratch to 69, as it would to trial factor from 69 to 70.

[diep]

Originally Posted by lavalamp

If it takes you 18 hours to do 68 -> 69, then it will take you 36 hours to do 69 -> 70, for a total of 54 hours.

Each bit increase represents a DOUBLING of the work done in the previous level.

To put it another way, it would take the same amount of CPU time to trial factor from scratch to 69, as it would to trial factor from 69 to 70.

The break even points i have here from GIMPS, which we also tried to use initially for Wagstaff, they are based upon old hardware.

Todays hardware is far faster in floating point and relative slower in trial factoring, so the break even points are not so accurate anymore.

"slower" i mean: the intels are ugly slow in trial factoring versus the AMD's very fast, yet in floating point both intel as well as todays AMD's are real fast for the LL. Something like moving from effectively 1-2 Gflops per cycle to 4+ today, versus trial factoring still same speed.

Vincent