Thinking out loud about getting under 20M unfactored exponents

[masser]

Originally Posted by VBCurtis

I've started ECM on the suggested range. My first set is 5 curves at B1=250K; I may adjust that later. A curve at this size on the Surface tablet I'm using takes about 25k seconds, so ought to progress at roughly 5 candidates per week. I'll give it a couple weeks to start.

Thanks again for the kind offer of help. I've attached a P-1 worktodo.txt for the 14.01M range if the ECM work becomes too boring. I'll ensure that my devices avoid that range for a few months (at least until mid-January).

[VBCurtis]

I accept your work!
I may inflate the bounds a little bit, after I time the first test or two.
"optimal" P-1 is out the window when we're going to need so much ECM to hit the factors-found target...

[VBCurtis]

I went with B1=8M, B2=200M for the first few P-1 jobs.
It's on a really slow but 8GB laptop, though I expect a desktop core to open up later.

[petrw1]

Since last update (July 24)
17 more ranges cleared: 4.0, 7.9, 8.8, 19.8, 22.5, 27.6, 29.4, 37.0, 37.3, 40.3, 40.5, 40.8, 41.0, 41.1, 45.5, 56.8, 59.4
And 4 bonus ranges: 68.4, 73.1, 73.5, 75.8.

There is only 1 bonus range to go (58.7)... and it had good progress; maybe; just maybe.
If or when it is cleared I am confident that every range 50M+ will be taken care of by natural GIMPS processes.
That allows this project to focus on the lower ranges only.

TOTALS to date:
195 total ranges cleared or 39.24%
22 Ranges with less than 20 to go.
1,601 more factored (25,598)....46.35% total factored.

Continuing to get lots of great help. THANKS

Thanks again for everyone contributing.

[petrw1]

Note that while there is lots of room for more TF in the remaining ranges, the lower the exponents the more beneficial P-1 becomes (or ECM for VERY low) and the less beneficial TF will become.

My general process is:
1. TF a couple extra bits, while they are still relatively fast for GPUs
2. Aggressively P-1 where the current B1=B2. (I choose new B1/B2 that give about a 2.5-3% improvement over the current P-1 You can use this tool: https://www.mersenne.ca/prob.php)
3a. Aggressively P-1 where the current B1/B2 are relatively low. (Any that I can gain at least 2% with reasonable new B1/B2)
3b. TF another bit level.
...(3a and 3b are numbered as such because the order I do these depends on analysis of which I estimate will help with the least effort/time. Often I'll do some of both).

That said; if you have TF-grade GPUs the immediate need is:
- TF the 20-29M ranges to at least 72 bits (or until a range is under 2,000 remaining)
===> If you use GPU72, it is already making available these assignments
- IF you have a more powerful GPU (or lots of them) consider helping in the more stubborn 30M and 40M ranges where aggressive P-1 is already done.
===> In the 40's: 44.6, 46.3 and 47.8 (They should be on GPU72 shortly)
===> In the 30's: 30.2, 31.6, 32.5, 33.0, 38.0, 38.6, 38.9
===> I (and others) are finishing P-1 for about 1 or 2 ranges a month to add to the above list.
- If you have CPUs or LL/P1-Grade GPUs consider doing aggressive P-1 as I defined above (or ECM if your RAM is limited). Pick a range and reserve it HERE and/or here: https://www.mersenneforum.org/showth...hlight=redoing

Thanks in advance.

[KEP]

@Everyone:

Due to the difficulty of predicting Colab availability, I've decided to still help this effort and work far far far away from where SRBase is crunching and climb 1 bit at a time. The choice fell on the range 999M-1G (1 million n).

Currently I've loaded all exponents Trial Factored to 72 bit, to my worktodo.txt file. If I can complete about 100 a day, it will take me about 107 days using Colab, to completely test 72 to 73 bit. Once that completes, I'm going to load all candidates Trial Factored to 73 bit and take them to 74 bit.

My hope is that someone gives me a heads up, if they reserve any large amount of work, so I can remove the candidates from my list. Even though Colab ressources are free, it is just not meaningfull to run 2 trial factor tests of the same n's. Also I will keep an eye on the Active Assignments page and see if anyone reserves anything for Trial Factoring.

Take care.

[lycorn]

Originally Posted by KEP

My hope is that someone gives me a heads up, if they reserve any large amount of work, so I can remove the candidates from my list.

Why don't you simply reserve the batches you're willing to test, like the SRBASE folks do?

[KEP]

Originally Posted by lycorn

Why don't you simply reserve the batches you're willing to test, like the SRBASE folks do?

Good and just question. As far as I know, there is no way for me to reserve all candidates tested to 72 bit and leave those tested to higher bit levels unreserved. I either have to make potentially thousands of reservations to get just those tested to 72 bit and none of those tested to higher bits. I tried using the manual GPU reservation page a while back and it gave me 100 n's in the low n range, near the wavefront. If there is a way for me to reserve for 999M-1G all candidates tested to 72 bit please let me know, either here or in PM

[masser]

Originally Posted by KEP

Good and just question. As far as I know, there is no way for me to reserve all candidates tested to 72 bit and leave those tested to higher bit levels unreserved. I either have to make potentially thousands of reservations to get just those tested to 72 bit and none of those tested to higher bits. I tried using the manual GPU reservation page a while back and it gave me 100 n's in the low n range, near the wavefront. If there is a way for me to reserve for 999M-1G all candidates tested to 72 bit please let me know, either here or in PM

Is this what you seek?

[lycorn]

Originally Posted by masser

Is this what you seek?

That link allows you to get exponents to test, but it doesn´t actually reserves them in the server.

But you can do it using the Manual GPU Assignments Form:

i) Choose the number of assignments you wish (don´t know whether or not there is a limit)
ii) Leave the Preferred Work Range at "What makes sense"
iii) Enter 999000000 and 999999999 in the Optional Exponent range fields
iv) Leave Work Preference at "What makes sense"
v) In the "Optional bit level to factor to" field enter 73. Note: I tried it and it gave me assignments from 72 to 74 instead. It´s not a problem, tho, just a matter of editing the worktodo file changing ",74" to ",73". Notepad will suffice.

Even if there are limits on the number of exponents you may reserve, that´s not a problem. After all, you don´t really need to reserve all available exponents in one go, do you?

HTH

[KEP]

Originally Posted by masser

Is this what you seek?

Not exactly, but it was a good shoot - as Lycorn mentions, it did not give me the possibility to reserve anything, but it was very usefull for my initial plan, since it was just copy paste and then push "play" in the notebook.

Originally Posted by lycorn

That link allows you to get exponents to test, but it doesn´t actually reserves them in the server.

But you can do it using the Manual GPU Assignments Form:

i) Choose the number of assignments you wish (don´t know whether or not there is a limit)
ii) Leave the Preferred Work Range at "What makes sense"
iii) Enter 999000000 and 999999999 in the Optional Exponent range fields
iv) Leave Work Preference at "What makes sense"
v) In the "Optional bit level to factor to" field enter 73. Note: I tried it and it gave me assignments from 72 to 74 instead. It´s not a problem, tho, just a matter of editing the worktodo file changing ",74" to ",73". Notepad will suffice.

Even if there are limits on the number of exponents you may reserve, that´s not a problem. After all, you don´t really need to reserve all available exponents in one go, do you?

HTH

It appears that I must have done something wrong, on the last try of reserving for manual Trial Factoring assignments ahead of the wavefront. The descriped plan worked, however now I have changed my own plan and decided to Trial Factor from 72 to 74 bit on the currently 2000 reserved tasks.

You are right, I do not need to reserve all work at once