mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > Aliquot Sequences

Reply
 
Thread Tools
Old 2022-07-13, 23:06   #1750
gd_barnes
 
gd_barnes's Avatar
 
May 2007
Kansas; USA

261148 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
For comparison:
I have recently acquired a Ryzen 5950X. Running at factory speed with DDR4-3200 memory, I can factor a C135 in 4 hours. I haven't yet attempted a C150, but it projects to take 24-28 hours on that 16-core CPU.

A 64-core machine is likely to run at around 2/3 the Ghz, something around 2.5x the speed of my 5950X. So, I would estimate a C155 to take on the order of 20-25 hours. Your timing data seems fine, then.

It's notable that it is now under a day for a (large, expensive) single machine to crack numbers of 512 bits size.
Interesting.

Do your times above include ECM time? Or are they just NFS time or whatever equivalent is used in CADO? How much memory for a C135? About how much memory for a C155?

I've had a Ryzen 3950X (16 cores/32 threads) for ~18 months and just recently fired it up on factoring although I believe I've done nothing bigger than a C123. In the near future, I'd like to give it a crack at a C137. I'd like to do some comparison.

I'm running Windows 10 so I don't expect to be as fast plus the 5950X is certainly going to be faster than a 3950X. I'm curious to do some extrapolation.
gd_barnes is online now   Reply With Quote
Old 2022-07-13, 23:26   #1751
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

13×421 Posts
Default

I'm away on vacation for a couple more days, so I don't have precise timing data handy. The time I cited is CADO-NFS claimed wall clock time for a factorization, so ignores ECM. Naturally, one can choose how much ECM to do; I choose about 20% of NFS time for my ECM effort as a rough guideline. I'll have a look at 'top' in linux while a job is running to discover memory use within CADO- I think part of post processing uses as much as 15GB even in the C135-C140 range, but it's likely there's a lower-memory option CADO chooses if tons of memory isn't available.

I'm taking the opportunity of a new CPU to redo my CADO params files; I'm finding 10-15% improvement over my old files at most sizes, basically one digit faster. I'll get the smaller ones posted in a week or two after some more testing.

If there are others still using factmsieve.py for parameter selection rather than YAFU, I may try to improve parameter selection for 100-130 digit jobs using what I've learned from CADO param selection; mostly that's much much larger large-prime choices than the old rules of thumb (I'm at 29 bit LP by C135, for instance, even while still on 13e). I don't know how to tell YAFU what to choose, but perhaps if I played with factmsieve the improved settings might be imported into YAFU? I mean, I think 25% faster jobs are possible on the GGNFS sievers.

Last fiddled with by VBCurtis on 2022-07-13 at 23:27
VBCurtis is offline   Reply With Quote
Old 2022-07-14, 00:42   #1752
charybdis
 
charybdis's Avatar
 
Apr 2020

32·5·19 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
I think part of post processing uses as much as 15GB even in the C135-C140 range, but it's likely there's a lower-memory option CADO chooses if tons of memory isn't available.
I think it's just the square root that can hit such high memory at this range, as if you don't set tasks.sqrt.threads it will run as many dependencies as threads. As you know, filtering and linear algebra cause memory issues for much larger numbers and I don't think there's any way around this apart from using msieve instead.
charybdis is offline   Reply With Quote
Old 2022-07-15, 03:16   #1753
gd_barnes
 
gd_barnes's Avatar
 
May 2007
Kansas; USA

22×34×5×7 Posts
Default

Quote:
Originally Posted by gd_barnes View Post
Ongoing:
Continue on (B) to initialize all same-parity exponents on the project with beginning size of <= 180 digits. Remaining: Bases 76 to 199, 400 to 1000, and > 1250. I expect to be done up to base 100 later today and with all bases within ~ 3 days.
I forgot to report this a few days ago...I'm finished initializing all same-parity exponents on all bases with beginning size of <= 180 digits. It led to many terminations and there are still many more interesting sequences that have come out of it. :-)
gd_barnes is online now   Reply With Quote
Old 2022-07-15, 09:53   #1754
garambois
 
garambois's Avatar
 
"Garambois Jean-Luc"
Oct 2011
France

977 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
For comparison:
I have recently acquired a Ryzen 5950X. Running at factory speed with DDR4-3200 memory, I can factor a C135 in 4 hours. I haven't yet attempted a C150, but it projects to take 24-28 hours on that 16-core CPU.

A 64-core machine is likely to run at around 2/3 the Ghz, something around 2.5x the speed of my 5950X. So, I would estimate a C155 to take on the order of 20-25 hours. Your timing data seems fine, then.

It's notable that it is now under a day for a (large, expensive) single machine to crack numbers of 512 bits size.
I have serious problems with CADO-NFS when the "Info:Linear Algebra: lingen ETA: not available yet" part starts.
The processor is loaded to 100%, but the job does not run.
And I noticed that there is no such problem if I stop all other running calculations in parallel on the same computer.
Does anyone know if there is a setting or option to change somewhere to avoid this problem ?
Indeed, I usually process several sequences in parallel and also run several other programs on the aliquot sequences in parallel on my computer.
This problem for the "Linear Algebra: lingen" part prevents me from doing this and it is extremely annoying for me.
garambois is offline   Reply With Quote
Old 2022-07-15, 10:03   #1755
sweety439
 
"99(4^34019)99 palind"
Nov 2016
(P^81993)SZ base 36

72·73 Posts
Default

Quote:
Originally Posted by garambois View Post
On my website, I wrote a page that summarizes my work on aliquot sequences starting on integer powers n^i. This page summarizes the results and reservations for each aliquot sequences one has chosen to calculate.

See this page.
You forgot to add base 120.

Also, you can add a new category “factorials” (just like “primorials”), including bases 2, 6, 24, 120, and 720 (you can also consider base 5040)

Also, you can add a new category “Lehmer five”, including bases 276, 552, 564, 660, 966
sweety439 is offline   Reply With Quote
Old 2022-07-15, 10:23   #1756
garambois
 
garambois's Avatar
 
"Garambois Jean-Luc"
Oct 2011
France

977 Posts
Default

I have a question about the Sage software.
This is the software I am using to do my data analysis on this project.

Let's have a list of point coordinates, like this one for example (I imagined this list completely at random to make you understand what I am looking for):
Code:
[(1,1),(1,3),(1,6),(2,4),(2,5),(2,10),(2,33),(3,2),(3,9),(3,12),(3,45),(4,2),(4,10),(4,16),(5,2),(5,25),(5,60)]
My coordinate lists consist of only integers.

Is there an instruction on the Sage software that can find me automatically that several of these points are on a known curve that has a known equation.

You may have found the curve by looking at the data, because the coordinates [(1,1),(2,4),(3,9),(4,16),(5,25)] all define points that are on the parabola with equation y = x^2.
I'm not just looking for an instruction that would give me the polynomial type equations, but of any kind, something very general that would give me the points belonging to lines, curves of polynomial equations, exponential and whatever else is possible and known.
I have no idea how to program this myself.

Warning : I am not looking for an instruction to see if all points in the given set belong to a curve of known equation, but if a "small" subset of the given points belong exactly to a curve of known equation.

And if you don't know if this instruction exists for the Sage software, can you tell me if you know of another software that can do this job ?
garambois is offline   Reply With Quote
Old 2022-07-15, 10:53   #1757
garambois
 
garambois's Avatar
 
"Garambois Jean-Luc"
Oct 2011
France

977 Posts
Default

I am in the process of doing the big update.
I'm also fixing the yo-yo bookings that show up on the page but are no longer booked.
By the way, the matched parity sequences of the base 60 will be released by yoyo.
And the last sequences of this base, for exponents 86, 88 and 94 end with a small prime number.
I think it's a coincidence and that for exponents 90, 92, 96, 98 and 100, this will not be the case.
I can't wait to see the ending primes of base 60 !


Quote:
Originally Posted by sweety439 View Post
You forgot to add base 120.
This base has never been officially initialized.
But it seems that some work has been done on it.
Does anyone want to finish the initialization of this base 120 ?


Quote:
Originally Posted by sweety439 View Post
Also, you can add a new category “factorials” (just like “primorials”), including bases 2, 6, 24, 120, and 720 (you can also consider base 5040).
Why not, I had already thought of that.
OK, I'll do it.
That said, at the moment I don't notice any special behavior from the bases that are factorials.
As for the 120 and 5040 bases, I would only add them if someone initializes them.


Quote:
Originally Posted by sweety439 View Post
Also, you can add a new category “Lehmer five”, including bases 276, 552, 564, 660, 966
We had already discussed this a long time ago.
This category may not be final, so we will not add it.
garambois is offline   Reply With Quote
Old 2022-07-15, 13:13   #1758
EdH
 
EdH's Avatar
 
"Ed Hall"
Dec 2009
Adirondack Mtns

2·32·271 Posts
Default

Quote:
Originally Posted by garambois View Post
I have serious problems with CADO-NFS when the "Info:Linear Algebra: lingen ETA: not available yet" part starts.
The processor is loaded to 100%, but the job does not run.
And I noticed that there is no such problem if I stop all other running calculations in parallel on the same computer.
Does anyone know if there is a setting or option to change somewhere to avoid this problem ?
Indeed, I usually process several sequences in parallel and also run several other programs on the aliquot sequences in parallel on my computer.
This problem for the "Linear Algebra: lingen" part prevents me from doing this and it is extremely annoying for me.
I've noticed this and saved some logs a little while back, but I didn't pursue anything further because I use Msieve for LA. I'll try to dig out some info and contact the CADO-NFS Team.

I wonder if there's a priority (nice?) setting involved.

@ charybdis: Do you have any insight?
EdH is offline   Reply With Quote
Old 2022-07-15, 17:50   #1759
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

10101011000012 Posts
Default

I have the same problem- even one other thread running makes lingen take hours rather than minutes.
Even more strangely, this happens even if fewer total threads are running on a CPU than the number of cores it has. A 4-threaded CADO with 4 copies of ECM on a 10-core chip produces the same problem.
VBCurtis is offline   Reply With Quote
Old 2022-07-16, 11:03   #1760
garambois
 
garambois's Avatar
 
"Garambois Jean-Luc"
Oct 2011
France

97710 Posts
Default

Page updated.
Many thanks to all for your help.
Thank you very much for notifying me of any oversights or malfunctions.

Added bases : 179 (prime), 1210 (2-cycle) and 31704 (28-cycle).
Updated bases : all the bases of the project !
Corrections : correction of erroneous allocations for bases 239 and 241 (see post #1735).
Reservations : delete yoyo reservations for all bases for which yoyo has reached its limits according to this page.
Sort categories : added the new factorial sort category to the page (many thanks to Karsten Bonath for his help !).

The removal of yoyo reservations has freed up sequences of matched parity.
The summer work has started well.
I will do another full update in the middle of August before the big data harvest and certainly some smaller updates between now and then.

I would still like to emphasize the question I asked in post #1756.
It's a very important question and if anyone has a lead to give me, please don't hesitate to contact me.
garambois is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Broken aliquot sequences fivemack FactorDB 46 2021-02-21 10:46
Broken aliquot sequences schickel FactorDB 18 2013-06-12 16:09
A new theorem about aliquot sequences garambois Aliquot Sequences 34 2012-06-10 21:53
poaching aliquot sequences... Andi47 FactorDB 21 2011-12-29 21:11
New article on aliquot sequences schickel mersennewiki 0 2008-12-30 07:07

All times are UTC. The time now is 03:49.


Wed Sep 28 03:49:39 UTC 2022 up 41 days, 1:18, 0 users, load averages: 1.36, 1.26, 1.09

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔