Newer milestone thread

[lycorn]

Correct. And if you and I hadn´t helped, LaurV would probably be taking 2 or 3 more days...

[TheMawn]

Good idea, LaurV. If you need help, I could probably divert my laptop for a few days.

[blahpy]

In order to clear up another range I've quickly written up a small Java program to convert factoring effort reports into mfaktx lines, so I'll get a set of lines to clean up 800M - 900M to 66 (even though it's pretty pointless work, that range is really close to being done but the leftovers that are only TFed to 65 are sparse and OCD says it should be fixed).

If someone wants the lines when I have them (Mawn perhaps?), I could quite easily send them over, there are only 6,000 or so and they will take probably 5 or so seconds per exponent on a GTX 670 or whatever you guys use.

[LaurV]

Because people keep asking, I have asked Oliver about distributing the 0.18_low_expos version of mfaktc to third parties. He says like that (the "effort" of bolding few of the words is mine, he said them, but I took the liberty to stress them, because I very much agree):

Originally Posted by TheJudger

Technically it is GPL and you can do what you want with it. My advice is: use this version "with care", e.g. only on exponents which the official release version can't do. This is on my todo list because it was requested so many times. I'll add some testcases, do some runs with full debugging enabled and if it works I'll enable it for regular releases. If it doesn't work (without simple fixes) I'll ignore the requests for low exponents, I still think it is not really useful. ;)

So, here it is attached.

Be warned that: (or subjective motivation why it was not released to the public, beside of the other motivations explained before).

1. The program is SLOW. Still, few times faster than the CPU, but because of the technical/mathematical details explained so many times (sieving large primes for small expos, large candidate base for small expos, etc), do not expect to get the same performance like you would get with the "regular Mfaktc 0.20 working at the LL front". You will get about 1/20-1/50 of that performance, depending on your card. If you are "in" for the credit, this program is not for you (read the next point too: CPU performance is also hurt).

2. The program uses the CPU for sieving. You can use the ini files from the last version (0.20 or +, i.e. you don't need to create new ini files for it) to set the number of iterations, gpu number, if you have many, etc, but everything related to "gpu sieving" will be ignored. That part is developed long after this program was released. So, you will see the performance halving/doubling if you start/stop P95, or if your CPU is free/busy, etc. It is more suitable for when the CPU is free, or you want to keep it not very busy/hot for different reasons.

3. Due to the point above, the program may not maximize your card. This is good when you want to control the heat and power consumption, but will result in even lower credit for you. So, use GPU-Z to see how much your card is used, and if you want to squeeze the gpu, you will need to launch more copies of the program in the same card (bad part: each will use a CPU core; good part: logical core is enough). Put them in different folders, with its own ini, cudart (btw, see below), worktodo file, etc, and eventually use Misfit to control all those instances. You may need 3-4-5 instances to max a gtx580, depending on your CPU, and if you have 4 cards inside the box, it will be a hell to control 20 instances without Misfit. Here is where Misfit is the king, and that is why it was done. New kids here are spoiled with new mfaktc versions which can filter with the GPU and max a card alone, without using the CPU at all, so the utility of Misfit is lost somehow, but the things were not always the same. A batch file is in the zip, just in case you want to use it as startup or inside of misfit, but first you look inside and edit the paths/names to your folders. The "questioning" in the beginning is very helpful, don't delete it, or you may end up launching it multiple times, they will all work in the same folders, creating a mess... ("don't ask us how we know" (TM))

4. You will need the suitable cudart library. It says which one when you started it, you can dld the library from the forum's ftp, or from cudaLucas site, regular mfaktc sites, etc, there is nothing magic about the library. Put it in the same folder(s) as the program. [edit: I put one in the zip, but in case you don't trust it... (I was thinking two times before putting it, first because of the size, second because of the safety. I added a checksum file)]

5. As pointed by Oliver above, and by others many times (me included) you will not help GIMPS too much by working lowest_exponents. From GIMPS point of view, their status is cleared long time ago. You may be the cool guy to find a big factor for a small exponent, which may be same hard as finding a new mersenne prime () considering the ECM done for those expos. That would be cool indeed

6. Profit...


[Xyzzy: if you want , you can move the file in the right folders, to clean the space used by my quota]

[lycorn]

Thx a bunch.
Have already tried it. I trusted "your" library, and it worked fine.
Will put it to use one of these days, upon finishing some work currently underway.

[Uncwilly]

Getting the thread back on track to its intended porpoise.

All exponents below 26,572,589 have been tested and double-checked.
All exponents below 44,944,841 have been tested at least once.

Countdown to testing all exponents below M(57885161) once: 16,699

Countdown to proving M(30402457) is the 43rd Mersenne Prime: 462
Countdown to proving M(32582657) is the 44rd Mersenne Prime: 16,598

462 is a bit too much work to expect a "Christmas miracle" this year (or even a tidying up by the end of the year.)

[TheMawn]

462 DC before Christmas is super easy. We just need a few people on board. I'd be cool for doing a reserved exponent for no credit (and then giving whoever owns the assignment the credit when they submit the assignment IF they ever do) to clear up the ones which are probably being perpetually reserved.

[kracker]

Originally Posted by TheMawn

462 DC before Christmas is super easy. We just need a few people on board. I'd be cool for doing a reserved exponent for no credit (and then giving whoever owns the assignment the credit when they submit the assignment IF they ever do) to clear up the ones which are probably being perpetually reserved.

Indeed. I can do up to 100 DC a month, so I really don't think 462 is impossible by current whole output.

[TheMawn]

Hell, I'd even be up for giving them the credit.

Merry Christmas! I did you assignment for you and here's your credit. This would be a really good way to clear up these "stuck" assignments.

[philmoore]

Originally Posted by Uncwilly

All exponents below 26,114,633 have been tested and double-checked.
All exponents below 44,576,437 have been tested at least once.

Countdown to testing all exponents below M(57885161) once: 21,382

Countdown to proving M(30402457) is the 43rd Mersenne Prime: 924
Countdown to proving M(32582657) is the 44rd Mersenne Prime: 23,274

This quote is from October 3rd - I just want to point out that progress has been pretty steady on this first double-check milestone since then, and current GIMPS strategy has been to only hand out the lowest available double checks to "most trusted" participants. Rather than poach large numbers of assignments from long time participants who have worked diligently to obtain this "most trusted" status, why not work to make sure that all three of these milestones are reached some time in 2014, a very reasonable and attainable goal IMO? In a couple of months, it may be obvious that a handful of these assignments are indeed stuck, but why duplicate effort and possibly alienate some loyal contributors in the meantime?

[LaurV]

Originally Posted by TheMawn

462 DC before Christmas is super easy. We just need a few people on board. I'd be cool for doing a reserved exponent for no credit (and then giving whoever owns the assignment the credit when they submit the assignment IF they ever do) to clear up the ones which are probably being perpetually reserved.

The real problem is the hanged exponents. Under no circumstances will George or Chris encourage poaching them...
(edit, sorry, I see it was already addressed)