Thinking out loud about getting under 20M unfactored exponents

[pinhodecarlos]

Having lots of fun running P-1 with not the recommended optimized B1/B2 bounds but even though I am getting satisfying rate of factors for range below 7.8M. Work was taken from mersenne.ca at the "poorly P-1 blablabla" link.

[petrw1]

Originally Posted by pinhodecarlos

Having lots of fun running P-1 with not the recommended optimized B1/B2 bounds but even though I am getting satisfying rate of factors for range below 7.8M. Work was taken from mersenne.ca at the "poorly P-1 blablabla" link.

Cool... it all helps

[pinhodecarlos]

Originally Posted by petrw1

Cool... it all helps

Will go back to your 16M once I’m done with these, 10 days to go.

[petrw1]

Originally Posted by pinhodecarlos

Will go back to your 16M once I’m done with these, 10 days to go.

No Longer Required....someone else cleared the 16.6M range.

But when you are ready for other small ranges let me know.

Thanks

[petrw1]

Available now (P-1 done)
29.1
35.5
49.4

A little more P-1 is in progress
30.1
36.2

A little more P-1 would be beneficial but not essential
36.5
38.2

[firejuggler]

I was thinking about doing a PM1 B1=500e3 B2 =10e6 on the 27-27.1 range ... Starting at about the new year. is the range availlable?

[petrw1]

Originally Posted by firejuggler

I was thinking about doing a PM1 B1=500e3 B2 =10e6 on the 27-27.1 range ... Starting at about the new year. is the range availlable?

It is free and I absolutely appreciate any help. But if I may offer some experience:

In my humble opinion if you mean 27.0 that is a tough one. I'm not trying to talk you out of it, just want to be sure you know what you are getting into.
If you mean 27.1 that is much closer/easier.
Based on a lot of math I use to determine the best way to clear these ranges, I think your bounds could be higher, though.

HERE IS THE MATH I USE:
The 27.0 range needs 176 more factors. TF alone would take too much work..
Each bit of TF clears a little over 1% (of 2175 ... 25 would be generous); so we expect something like:
TF70-71: 2175-25=2150 (8.85 GhzDays per assignment)
TF71-72: 2150-25=2125 (17.7)
TF72-73: 2125-25=2100 (35.5)
TF73-74: 2100-25=2075 (71.0)
TF74-75: 2075-25=2050 (142.0)
I'm not sure we want to go this high so TF alone is not the answer.

So we need at least 50 factors from P-1; either before TF or after.

There are 860 that have had a low B1=B2; about 420e3
For these 860 your proposed bounds have an expected success rate of close to 3% or about 25 factors according to these:
https://www.mersenne.ca/prob.php?exp...0000&b2=420000
https://www.mersenne.ca/prob.php?exp...00&b2=10000000.

The remaining 1300 or so have current higher B1/B2 which would diminish your success rate if you continue to P-1 these as well.
Something like a 2% success rate here gives another 26 factors....we're getting closer. But I'm not sure we want to P-1 all 2175 exponents either.

On the other hand if you used bounds of 1e6,20e6 you'll get another 1% (granted for double the work; 2.7 vs 1.35 GhzDays each).
https://www.mersenne.ca/prob.php?exp...00&b2=20000000
So out of the 860 you would get about 34 factors.
And if your goal is 50 P-1 factors you need to P-1 a lot less of the remaining 1300 exponents.

On the third hand (hahaha), sometimes I will use the lower B1/B2 for the exponents that have a current B1=B2 and more aggressive B1/B2 for the others.
I have a spread sheet that helps me choose.

CONFUSED.....or maybe you are way better at stats than me and know all this and more.
Anyway thanks a lot and anything you choose is greatly appreciated.

[petrw1]

Originally Posted by petrw1

Available now (P-1 done)
29.1 .... I'm working on this one
35.5
49.4

A little more P-1 is in progress
30.1
36.2

A little more P-1 would be beneficial but not essential
36.5
38.2

I share. Use GPU72 to avoid toe stepping OR start at the top (i.e. 29100000)

[Prime95]

I'm looking for volunteers to test a pre-release of version 30.4. 30.4 is a nearly complete rewrite of the ECM and P-1 code.

Useful testing would include:
1) rerun successful P-1 and ECM attempts to make sure the new code does not miss any factors.
2) make sure save and restore work
3) find bugs reading old save files
4) testing various high and low memory configurations
5) report back on whether (how much) the new code is faster

and finally...

6) run the improved code to see if it is finding about the expected number of new factors.

Any interest? Please specify Windows or Linux.

New features include:
- better use of available memory for ECM and P-1 stage 2
- small speed improvements in ECM stage 1
- selection of optimal B2 in both ECM and P-1
- deprecate Brent Suyama (more efficient to put that effort into a larger B1)

[kruoli]

Wonderful! I'm looking forward to this. I could try Windows and Linux both, 64 bits. Thank you!

[chalsall]

Originally Posted by kruoli

Wonderful! I'm looking forward to this. I could try Windows and Linux both, 64 bits. Thank you!

I've currently got a couple of machines doing DC-P1 in 104M.

Would love to give it a whirl if you don't there's (much) risk of missed factors. (If a bug is found they can simply be re-run.)