mersenneforum.org Divergent Posts from Benchmark Thread
 Register FAQ Search Today's Posts Mark Forums Read

 2022-08-12, 14:30 #89 swellman     Jun 2012 71478 Posts I used msieve to find a polynomial for a c175. msieve can’t sieve - GGNFS must be used for sieving - one uses msieve to find a GNFS polynomial (or build your own in the case of SNFS) - sieve using GGNFS. This will take the most time in the entire factoring process. - use msieve to filter, run linear algebra, and finally run a square root routine to extract factors from the LA results. (The details are much more complicated and not my story to tell. Nope.) - sip your favorite cold beverage and bask in the satisfaction of a job well done. CADO can also do these steps with a “fire and forget” capability. EdH has laid out a nice process for setting up CADO. Yafu, which apparently you’ve used the past, also does everything for you. As @VBCurtis mentioned in an earlier post, it uses msieve, GGNFS, even GMP-ECM in its processes to factor a number. But msieve by itself can’t factor a number with NFS methods. You need GGNFS. Or CADO to sieve and the throw the results into msieve but that’s a discussion for another time.
2022-08-12, 14:59   #90
storm5510
Random Account

Aug 2009
Not U. + S.A.

2,333 Posts

Quote:
 Originally Posted by swellman I used msieve to find a polynomial for a c175. msieve can’t sieve - GGNFS must be used for sieving - one uses msieve to find a GNFS polynomial (or build your own in the case of SNFS) - sieve using GGNFS. This will take the most time in the entire factoring process. - use msieve to filter, run linear algebra, and finally run a square root routine to extract factors from the LA results. (The details are much more complicated and not my story to tell. Nope.) - sip your favorite cold beverage and bask in the satisfaction of a job well done. CADO can also do these steps with a “fire and forget” capability. EdH has laid out a nice process for setting up CADO. Yafu, which apparently you’ve used the past, also does everything for you. As @VBCurtis mentioned in an earlier post, it uses msieve, GGNFS, even GMP-ECM in its processes to factor a number. But msieve by itself can’t factor a number with NFS methods. You need GGNFS. Or CADO to sieve and the throw the results into msieve but that’s a discussion for another time.
I have the ggnfs library linked to my Yafu setup. I saw it being used on my assignment yesterday evening late after I started using the -v switch. It just made visible more detail about what was happening.

To tell the truth, I feel a bit deflated after all this struggling to learn msieve cannot do what I need. I will have to study your "CADO" link in more detail. It seems like a lot and some of it is a bit foggy to me.

 2022-08-12, 16:26 #91 EdH     "Ed Hall" Dec 2009 Adirondack Mtns 4,909 Posts Let me add some things to this mix: Running Msieve for those parts of the process it can be used for, is the only way I know of to make use of the GPU, which is advantageous. Using Msieve for LA, even without the GPU, is faster than the CADO-NFS LA*. There are other packages that make use of Msieve and GGNFS, and can use the GPU, but all of these will take a little study to make use of. I do have another "How I. . ." which uses one of them called factmsieve.py. These do not include any ECM as YAFU does. * CADO-NFS can also use the GPU for one stage during LA, but it is not an easy process to set up. I thought YAFU could use the GPU during Msieve, but can't find that in the documents. I've posed the question to the author. Perhaps someone reading this knows. Last fiddled with by EdH on 2022-08-13 at 12:19 Reason: I confused GPU with MPI
2022-08-12, 17:20   #92
frmky

Jul 2003
So Cal

32·52·11 Posts

Quote:
 Originally Posted by EdH I thought YAFU could use the GPU during Msieve, but can't find that in the documents. I've posed the question to the author. Perhaps someone reading this knows.
That should be just a matter of having all the appropriate files in the right place including the *.ptx and cub/*.so files, and passing the option "-g 0" to YAFU's LA msieve command.

2022-08-12, 17:33   #93
storm5510
Random Account

Aug 2009
Not U. + S.A.

44358 Posts

Quote:
 Originally Posted by EdH Let me add some things to this mix: Running Msieve for those parts of the process it can be used for, is the only way I know of to make use of the GPU, which is advantageous. Using Msieve for LA, even without the GPU, is faster than the CADO-NFS LA*. There are other packages that make use of Msieve and GGNFS, and can use the GPU, but all of these will take a little study to make use of. I do have another "How I. . ." which uses one of them called factmsieve.py. These do not include any ECM as YAFU does. * CADO-NFS can also use the GPU for one stage during LA, but it is not an easy process to set up. I thought YAFU could use the GPU during Msieve, but can't find that in the documents. I've posed the question to the author. Perhaps someone reading this knows.
The good thing about all this was leaving the Windows 10 SATA SSD intact in the Xeon. All I had to do was switch the cables.

After reading the quote part above, I think I will just stick with Yafu. That system won't run anything newer than v1.34. Participating in James' factoring project, I learned quite a bit more about Yafu. There is a way to set a tolerance on how far the ECM part goes, if I decide to use it. Mostly, I don't use it. I have it set to jump directly to NFS. I use 6 threads so it does alright that way. Running ECM can burn a lot of time.

For now, I will leave the Ubuntu drive intact. It seems to run really well, but my knowledge about it is very slim.

Edit/Update: I am curious about GGNFS. What I have is a library of binary files. Most have cryptic names. Is the reference made somewhere above something different?

Last fiddled with by storm5510 on 2022-08-12 at 18:33

2022-08-12, 19:53   #94
VBCurtis

"Curtis"
Feb 2005
Riverside, CA

2·2,741 Posts

Quote:
 Originally Posted by storm5510 Edit/Update: I am curious about GGNFS. What I have is a library of binary files. Most have cryptic names. Is the reference made somewhere above something different?
Those binaries are all there are for the GGNFS siever software. The number 11 through 16 in the binary names indicate the "size" of the siever, in that larger sievers are used to factor larger numbers. In GNFS jobs, each siever is best in a range of about 20-25 digits:
188+ uses 16
167-187 uses 15
142-167 uses 14
120-141 uses 13
etc
YAFU has such a map to choose which siever to use, with cutoffs close to the ones I've listed here. Near a cutoff (say, around 142 digits) it doesn't matter whether one chooses 13 or 14 to run the job- they will take about the same amount of time.

So, as long as you have the set of siever binaries where YAFU expects them to be, YAFU handles picking the settings for the siever and runs the siever. You can run the siever manually from the command line by looking at what YAFU invokes, and copy/pasting it to the command line. It's tedious, and not terribly helpful, but it can be done. I have this command in a text file that I copy/paste to the command line when I want to test-sieve a number; I try a range of parameter settings looking for the fastest ones before I run a whole job. YAFU has a set of "pretty good" parameters pre-set, and for small jobs (like under 150 digits) they're good enough.

It also seems that you don't look around this forum enough when someone references another program- there's an entire subforum for CADO-NFS, and the website for the software is listed literally adjacent to the subforum title (click on "factoring projects" folder and you'll see CADO-NFS as the second subforum, with the website link *right there* without even opening the subforum).

2022-08-12, 23:30   #95
storm5510
Random Account

Aug 2009
Not U. + S.A.

44358 Posts

Quote:
 Originally Posted by VBCurtis ...It also seems that you don't look around this forum enough when someone references another program- there's an entire subforum for CADO-NFS, and the website for the software is listed literally adjacent to the subforum title (click on "factoring projects" folder and you'll see CADO-NFS as the second subforum, with the website link *right there* without even opening the subforum).
I look around. This, for instance. Post #10. It just takes me longer, sometimes.

2022-08-15, 17:14   #96
storm5510
Random Account

Aug 2009
Not U. + S.A.

2,333 Posts

Quote:
 Originally Posted by frmky Did you remember the Code: git checkout update_cub
For some reason, this is no longer available.

Quote:
 error: pathspec 'update_cub' did not match any file(s) known to git

 2022-08-15, 19:34 #97 frmky     Jul 2003 So Cal 9AB16 Posts Sorry, I did some cleanup. It's now msieve-lacuda-nfsathome instead of update_cub.
2022-08-15, 23:05   #98
storm5510
Random Account

Aug 2009
Not U. + S.A.

2,333 Posts

Quote:
 Originally Posted by frmky Sorry, I did some cleanup. It's now msieve-lacuda-nfsathome instead of update_cub.
OK, well, whatever the case, I can't compile msieve.

Code:
make[1]: *** [Makefile:106: sort_engine.so] Error 1
make: *** [Makefile:360: cub/built] Error 2
git checkout msieve-lacuda-nfsathome

Code:
Already on 'msieve-lacuda-nfsathome'
Your branch is up to date with 'origin/msieve-lacuda-nfsathome'.
There must be something different...

 2022-08-16, 02:52 #99 frmky     Jul 2003 So Cal 9AB16 Posts If you are using CUDA toolkit version 11.5 or earlier, then Code: git checkout msieve-lacuda-nfsathome-cuda11.5

 Similar Threads Thread Thread Starter Forum Replies Last Post Xyzzy Hardware 877 2022-09-18 19:58 chalsall Forum Feedback 43 2019-03-26 16:10 Oddball Riesel Prime Search 5 2010-08-02 00:11 axn Sierpinski/Riesel Base 5 25 2010-05-28 23:57 masser Forum Feedback 3 2008-06-06 19:34

All times are UTC. The time now is 18:57.

Sun Oct 2 18:57:24 UTC 2022 up 45 days, 16:25, 0 users, load averages: 1.40, 1.44, 1.33