[QUOTE=Prime95;577308]I've taken 312000 to 312500. Running P-1 to 15M[/QUOTE]

Continuing on. Now doing 312500 to 314000 (exponents with known factors). P-1 with B1 = 20M.

When I tell prime95 to assume that TF has been done to 2^84 (the full 50K ECM level is complete), prime95 reports a little over 3% chance of finding a factor. I switched to 20M because I get a more traditional B2/B1 ratio there (not sure if that means switching from B1=15M to B1=20M was the right thing to do).

I have been using a slow laptop to work through a list of P+1 assignments that was previously posted. Can someone post some worktodo lines for P+1? The current task I have are in the 14M range. I would prefer some smaller exponents using a smaller FFT size. Currently the FFT size is 720K. A 256K FFT is about 3 times faster for me.

[QUOTE=Uncwilly;577403]I have been using a slow laptop to work through a list of P+1 assignments that was previously posted. Can someone post some worktodo lines for P+1? The current task I have are in the 14M range. I would prefer some smaller exponents using a smaller FFT size. Currently the FFT size is 720K. A 256K FFT is about 3 times faster for me.[/QUOTE]

Sorry about that. Thank you for the runs in the 14.0M range.

I've attached another file for the 1.6M range; no one has worked on the exponents with no factor in this range for a while, so these should be safe. They also will have much smaller FFT lengths. I also employed George's idea above to set the trial factoring depth to 84 to reflect the amount of ECM that has been performed on these exponents.

Let me know if these work; if not, I can try to generate another file for you.

[QUOTE=masser;577404]Sorry about that. Thank you for the runs in the 14.0M range.[/QUOTE]No need to apologize. You had no idea that my laptop is sluggish. It should be faster, but for some reasons it does not match the performance it should. I am happy to help with the exponents that I grabbed earlier. Exploring the new test is cool. I pulled this machine off of CF-PRP-DC to do this. That is about all this machine is useful for.

[QUOTE=Prime95;577399]When I tell prime95 to assume that TF has been done to 2^84 (the full 50K ECM level is complete), prime95 reports a little over 3% chance of finding a factor. I switched to 20M because I get a more traditional B2/B1 ratio there (not sure if that means switching from B1=15M to B1=20M was the right thing to do).[/QUOTE]

Full 50k ECM means there is a 1/e ~ 37% risk we missed a 25 digit factor if there is one, not quite the same as TF to 2^84 (26 digits).


[QUOTE=Uncwilly;577405]No need to apologize. You had no idea that my laptop is sluggish. It should be faster, but for some reasons it does not match the performance it should. [/QUOTE]

It probably throttles, you can try to use ThrottleStop which locks it at a Multiplier (frequency) you want. But be careful and watch the temperature in the beginning, so it does overheat. The program shows the temperature as well and you can set alarms:
[url]https://www.techpowerup.com/download/techpowerup-throttlestop/[/url]

[QUOTE=ATH;577423]Full 50k ECM means there is a 1/e ~ 37% risk we missed a 25 digit factor if there is one, not quite the same as TF to 2^84 (26 digits).[/QUOTE]


In my "optimal worktodo"-script, I calculate like the following (see [URL="https://github.com/keisentraut/prime95-optimal-worktodo/blob/master/get_work.py#L333"]around line 333[/URL]). I just use the current ECM bound, multiply by log2(10), round down and substract 12 extra bits. Only if this is larger than the TF bound, then how_far_factored will be increased. This usually only happens at very small exponents (<100k). The 12 bits are just gut feeling, I think the chance to miss a factor 12 bits less than the current ECM bound is <10%. For a full t30 ECM-factored number, how_far_factored will be set to at least 87 bits / 26 digits.
[CODE]
# use ECM bounds to adapt how_far_factored
# as ECM is probabilistic, we want to be conservative and remove an extra 12 bits / 4 digits of factor size
ecm_level = get_ecm_level(ecm)
how_far_factored = max(how_far_factored, int(ecm_level*math.log2(10)) - 12)[/CODE]So, please give it a try :) All you need to do is to install Python, download it from [URL="https://github.com/keisentraut/prime95-optimal-worktodo/"]github[/URL] and then run "python.exe get_work.py 123000 124000". Here is an example of it in work. Please note that the script decided that P-1 is already sufficient for this exponent, but P+1 should be done. If you don't agree with the strategy of this script, just tell me and I will adapt it. I'm not an expert either. And please excuse my spamming, but I think my script is much more convenient / efficient than just doing P-1 to a fixed B1 bound for a range.
[CODE]

# increased how_far_factored from 66 to 87 because of substantial ECM
# n: 77323
# how_far_factored: 87
# Factors: {11460140339503, 773231}
# factors known: True
# ECM Factoring: {(1000000, 100000000): 204, (250000, 25000000): 585, (50000, 5000000): 274, (11000, 1100000): 30}
# ECM level: t30
# ECM current B1: 1000000
# P-1 Factoring: {(100000000, 2000000000, 12)}
# P+1 Factoring: set()
# assigned: False
# fully factored: False
#
# should not do P-1: B1=100000000 recommended but 100000000 already done with B2=20.0*B1
Pplus1=N/A,1,2,77323,-1,50000000,0,1,87,"11460140339503,773231"
[/CODE]

[QUOTE=keisentraut;577507]The 12 bits are just gut feeling, I think the chance to miss a factor 12 bits less than the current ECM bound is <10%. For a full t30 ECM-factored number, how_far_factored will be set to at least 87 bits / 26 digits.[/QUOTE]

It is probably even less than 10% for 12 bits. I did some ecm tests years ago on constructed numbers with different size factors:
[url]https://mersenneforum.org/showpost.php?p=470180&postcount=48[/url]

You can see that a full T30 (30 digits = 10[SUP]29[/SUP]-10[SUP]30[/SUP] (96.3-99.6 bits)) finds factors:
1*10[SUP]28[/SUP] - 2*10[SUP]28[/SUP] (29 digits 93.0-94.0 bits): 79% of the time
1*10[SUP]27[/SUP] - 2*10[SUP]27[/SUP] (28 digits 89.7-90.7 bits): 88% of the time
1*10[SUP]26[/SUP] - 2*10[SUP]26[/SUP] (27 digits 86.4-87.4 bits): 97% of the time
1*10[SUP]25[/SUP] - 2*10[SUP]25[/SUP] (26 digits 83.0-84.0 bits): 99.6% of the time
1*10[SUP]24[/SUP] - 2*10[SUP]24[/SUP] (25 digits 79.7-80.7 bits): 100% of the time

You could probably subtract just 6-8 bits.

[QUOTE=keisentraut;577507]So, please give it a try :) All you need to do is to install Python, download it from [URL="https://github.com/keisentraut/prime95-optimal-worktodo/"]github[/URL] and then run "python.exe get_work.py 123000 124000".[/QUOTE]


Python newbie here. I downloaded this and tried to run on a Windows machine. Get an error message:

[CODE] File "C:\Users\woltm\Downloads\prime95-optimal-worktodo-master\get_work.py", line 6, in <module>
import urllib3
ModuleNotFoundError: No module named 'urllib3'[/CODE]

Ideas? I can run this on a Linux machine instead, but not as convenient for me.

[CODE] File "C:\Users\woltm\Downloads\prime95-optimal-worktodo-master\get_work.py", line 6, in <module>
import urllib3
ModuleNotFoundError: No module named 'urllib3'[/CODE]I missed that, was already installed on my Linux computer and I only tested this on Linux so far. Please download my script again because I have removed an unused dependency to beautifulsoup4, too. In order to install urllib3 in Windows, you need to use pip, the Python package manager.

See the stuff which I have added to the README.md:
[QUOTE]
[LIST][*]Download Python3 from the official website at: [URL]https://www.python.org/downloads/windows/[/URL][*]Install Python3 with default options[*]Use pip.exe (usually located in C:\Users\{USER_NAME}\AppData\Local\Programs\Python\Python{XX}\Scripts\) in order to install the requirement urllib3: pip.exe install urllib3[*]Run the script with python.exe get_work.py 123000 124000 1[/LIST][/QUOTE]

[QUOTE=masser;577404]Sorry about that. Thank you for the runs in the 14.0M range.

I've attached another file for the 1.6M range; no one has worked on the exponents with no factor in this range for a while, so these should be safe. They also will have much smaller FFT lengths. I also employed George's idea above to set the trial factoring depth to 84 to reflect the amount of ECM that has been performed on these exponents.

Let me know if these work; if not, I can try to generate another file for you.[/QUOTE]


Would you be kind to generate a range for me between 1.7M and 1.8M? TIA

[QUOTE=pinhodecarlos;577545]Would you be kind to generate a range for me between 1.7M and 1.8M? TIA[/QUOTE]


Please see attached. After reading comments above, I adjusted TF level to 77. Let me know if you need something different.