mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data > Marin's Mersenne-aries

Reply
 
Thread Tools
Old 2022-08-05, 17:59   #12
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

25·3·71 Posts
Default

DoubleCheck=332196607,81,1 assigned ll 2021-02-13, last update 2021-02-13, zero progress
DoubleCheck=332209313,81,1 mismatched res64s; Madpoo has TC assignment
DoubleCheck=332216777,81,1 assigned ll 2020-11-14, last update 2020-11-14, zero progress
DoubleCheck=332242709,81,1 assigned Kriesel 2022-01-15, ETA March 2023
DoubleCheck=332253617,81,1 assigned Kriesel 2022-01-15, ETA January 2023
DoubleCheck=332341057,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=332430691,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=333043493,81,1 assigned Kriesel 2022-01-15, ETA December 2023
DoubleCheck=333044087,81,1 mismatched prime95 res64s, Kriesel running LLTC on gpuowl, ETA 2022-08-22 (won't assign as LL, backup PRP assignment acquired)
DoubleCheck=340830817,84,1 won't manually assign, but claimed by Kriesel 2022-05-20, ETA 2022-08-16

Last fiddled with by kriesel on 2022-08-05 at 18:11
kriesel is online now   Reply With Quote
Old 2022-08-09, 06:19   #13
phillipsjk
 
Nov 2019

4216 Posts
Default

I am considering adopting 332216777 when my "white elephant" machine comes back from storage.


I was looking at the assignment rules, the exponents no longer expire if you selected "manual testing". What assurances do we have that the user "arpcar" has not been diligently working away on it offline for 2 years (other than the 360 day "time limit" that I may not be able to meet).


Which brings me to the "LL first test" requirements. My computer (a quad 16 core AMD Opteron(TM) Processor 6272 system, that I have been running at a reduced clock speed and 30% duty cycle) does not seem to meet the "Minimum Pentium 4 equivalent speed for a proven reliable computer -- 5,500 MHz" Requirement. Instead, the speed on PrimeNet is listed as: "1.399 GHz (0.746 GHz P4 effective equivalent)."
Is that multiplied by the number of cores at all, for granting assignments? (I would be dedicating 1 CPU or 16 cores -- anticipating around a year's run-time, given the 30% duty cycle and the 100 day 12 core estimate).


The computer certainly meets the reliability requirements (Reliability, Confidence: 0.99, 8.0). It is in storage because the wiring of the premises is unreliable and needs to be replaced (I power factor corrected LED lights in the bathroom by replacing half with the old incandescent bulbs: went from 0.74 to .95, reducing flickering).

Last fiddled with by phillipsjk on 2022-08-09 at 06:30 Reason: Punctuation tweaks
phillipsjk is offline   Reply With Quote
Old 2022-08-09, 09:26   #14
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
"name field"
Jun 2011
Thailand

100111011011012 Posts
Default

Hey Ken, let's trade!

We TC your mismatch 333044087 and you TC our mismatch 332328973 (we have the residues from both gpuOwl LL and cudaLucas, in case you need comparison on the way, what a pity they are not comparable to each-other - owl's residues' count is shifted by one or two, and therefore not useful, and Mihai doesn't seem interested in supporting LL anymore, i.e. align the residues properly with P95/cudaLucas/mlucas and implement shifts, etc). As the 0-shift run (with gpuOwl) was already done for our exponent, you may have to use the slower way (cudaLucas or P95 - IIANM, I think you squeeze a Phy anyhow?).
So, I can owl yours in 4-7 days (depending of how giving is gugu with the cards).

Deal?
LaurV is offline   Reply With Quote
Old 2022-08-09, 12:38   #15
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

152408 Posts
Default

Quote:
Originally Posted by phillipsjk View Post
I am considering adopting 332216777 when my "white elephant" machine comes back from storage.
Go for it. There's very little activity in 100Mdigit currently, judging by a recent review of https://www.mersenne.org/report_recent_results/ (27 TF, 1 P-1, 0 primality tests reported)


Quote:
Originally Posted by LaurV View Post
Hey Ken, let's trade!

We TC your mismatch 333044087 and you TC our mismatch 332328973 ...

Deal?
I've already started the 333044087 TC. Stay tuned, may take you up on it later.
In my experience gpuowl does a good job of matching Mlucas interim residues and vice versa. It's prime95/mprime that is an outlier, with its off-by-two iteration count.

See https://www.mersenneforum.org/showpo...4&postcount=12

https://www.mersenneforum.org/showpo...6&postcount=14
kriesel is online now   Reply With Quote
Old 2022-08-12, 05:19   #16
phillipsjk
 
Nov 2019

2·3·11 Posts
Default

Quote:
Originally Posted by kriesel View Post
Go for it. There's very little activity in 100Mdigit currently, judging by a recent review of https://www.mersenne.org/report_recent_results/ (27 TF, 1 P-1, 0 primality tests reported)

Reservation attempt failed:


Quote:
ra: already assigned, exponent:332216777, A: 1, b: 2, c: -1
No rush I suppose. Can't begin work on it until September.


Edit: looks like it was ready to run it anyway: had an assignment number of "N/A".


Incidentally, my slower computer is considered faster by PrimeNet. Maybe because of poor Interprocess Communication for the P-1 factoring ["White Elephant"] was doing (each core is about 37% idle). Only learned every core (less the main thread) was partly idle while completing work on my fileserver. FreeBSD top has a feature that shows the per core load when you hit the <F1> key.

Last fiddled with by phillipsjk on 2022-08-12 at 05:24 Reason: looked at worktodo
phillipsjk is offline   Reply With Quote
Old 2022-08-12, 14:29   #17
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

25×3×71 Posts
Default

Quote:
Originally Posted by phillipsjk View Post
Reservation attempt failed:
Yes; arpcar has had the assignment since November 2020, nearly 2 years, with zero progress reported.
kriesel is online now   Reply With Quote
Old 2022-08-18, 18:22   #18
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

25×3×71 Posts
Default status update

These are intended to be LL DC, TC, QC as needed to provide more data on LL reliability at 100Mdigit.

DoubleCheck=332196607,81,1 assigned ll 2021-02-13, last update 2021-02-13, zero progress
DoubleCheck=332209313,81,1 mismatched res64s; Madpoo has TC assignment, 0.1% done
DoubleCheck=332216777,81,1 assigned ll 2020-11-14, last update 2020-11-14, zero progress
DoubleCheck=332242709,81,1 assigned Kriesel 2022-01-15, 32.8% done, ETA March 2023
DoubleCheck=332253617,81,1 assigned Kriesel 2022-01-15, 34% done, ETA January 2023
DoubleCheck=332341057,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=332430691,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=333043493,81,1 assigned Kriesel 2022-01-15, 32.6% done, ETA December 2023
DoubleCheck=333044087,81,1 mismatched prime95 res64s, Kriesel running LLTC on gpuowl, 64.6% done, ETA 2022-08-23 (won't assign as LL, backup PRP assignment acquired)
DoubleCheck=340830817,84,1 DC did not match suspect first test; LLTC is assigned and in progress
kriesel is online now   Reply With Quote
Old 2022-08-19, 06:23   #19
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
"name field"
Jun 2011
Thailand

10,093 Posts
Default

Quote:
Originally Posted by kriesel View Post
In my experience gpuowl does a good job of matching Mlucas interim residues and vice versa. It's prime95/mprime that is an outlier, with its off-by-two iteration count.
Nope. Not entirely true.
We (royal we) didn't use MLucas up to now, but P95 can show all 3 residues all the time, so you have the +1 and +2 residues too, to match with any other mentioned tool, either gpwOwl or cudaLucas, that we DO use. So, cudaLucas always matched residues with (one of the) P95 ('s 3 residues). A small script, or 30 seconds editing of the log file with a clever text editor (like Notepad++, with regex search and replace) solves the issue of which one, and purge the rest from the output files, when you want to do comparisons.

In the past we used to run cudaLucas to both LL and DC for the exponents we run both, or sometimes P95 and cudaLucas, when we had the CPU resources available. GPU resources were always available, as we had lot more GPU power than CPU power, due to other things we use GPUs for.

However, since gpuOwl became MUCH faster, we try to combine it with cudaLucas, when running both LL and DC for the same exponent (and please don't start about why it is not good to run both LL and DC by the same user!).

Combining is needed in spite of the fact that cudaLucas is only (about) half-fast (therefore a lot of time is wasted using it, and if you do this on Colab, money are wasted too!), because gpuOwl doesn't have a random shift, so two results from gpuOwl won't be accepted by the server (which is the right action, and the fault of gpuOwl!).

However gpuOwl won't like to match those checkpoint residues with cudaLucas, at least not the version I have (compiled by you, by the way, and shared to me long time ago). We have many examples where the final residue matched, but none of the checkpoints did. We kept the logs, and can share them to whoever is interested.

Quote:
Originally Posted by kriesel View Post
<links>
TL;DR those interminable posts, a lot of ballast for too less cargo.

Last fiddled with by LaurV on 2022-08-19 at 06:30
LaurV is offline   Reply With Quote
Old 2022-08-19, 13:53   #20
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

11010101000002 Posts
Default

Quote:
Originally Posted by LaurV View Post
Nope. Not entirely true.
We (royal we) didn't use MLucas up to now, but P95 can show all 3 residues all the time, so you have the +1 and +2 residues too, to match with any other mentioned tool, either gpwOwl or cudaLucas, that we DO use. So, cudaLucas always matched residues with (one of the) P95 ('s 3 residues). A small script, or 30 seconds editing of the log file with a clever text editor (like Notepad++, with regex search and replace) solves the issue of which one, and purge the rest from the output files, when you want to do comparisons.
For clarity, & for those of us comparing by eye not script, prime95/mprime uses nonstandard residue iteration numbering high by 2, beginning with 2 and ending with exponent, rather than beginning with iteration 0 for the seed value 4, and ending at iteration # = exponent - 2, as the LL test is defined and performed in most other software capable of it. So a user must match the prime95 residue that is misnumbered iteration x+2 to another app's iteration #x (where x is typically a multiple of a power of ten). Prime95 is calling the displayed residues iteration x, x+1, x+2 in its interim residue output, but by standard numbering they are x-2, x-1, x respectively.

Quote:
gpuOwl won't like to match those checkpoint residues with cudaLucas, at least not the version I have ... We have many examples where the final residue matched, but none of the checkpoints did. We kept the logs, and can share them to whoever is interested.
I haven't seen that. I would be interested in seeing an example, perhaps posted as attachments. CUDALucas and gpuowl LL interim 64-bit residues matched in the rare cases I've checked when both runs were correct. The ~gigabit comparison here for example. (I generally no longer run CUDALucas because of its lower speed and lower reliability on the same hardware and task.)
Quote:
TL;DR those interminable posts, a lot of ballast for too less little cargo.
Yes, I recall you posting previously about your own tendency to do that. ;)
kriesel is online now   Reply With Quote
Old 2022-08-24, 04:54   #21
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

25×3×71 Posts
Default

These are intended to be LL DC, TC, QC as needed to provide more data on LL reliability at 100Mdigit.

DoubleCheck=332196607,81,1 assigned ll 2021-02-13, last update 2021-02-13, zero progress; kriesel running ETA ~Sep 8 2022
DoubleCheck=332209313,81,1 mismatched res64s; Madpoo has TC assignment, 0.1% done
DoubleCheck=332216777,81,1 assigned ll 2020-11-14, last update 2020-11-14, zero progress; phillipsjk may start in Sep 2022
DoubleCheck=332242709,81,1 LLDC assigned Kriesel 2022-01-15, 33.4% done, ETA March 2023
DoubleCheck=332253617,81,1 LLDC assigned Kriesel 2022-01-15, 35.5% done, ETA January 2023
DoubleCheck=332341057,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=332430691,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=333043493,81,1 LLDC assigned Kriesel 2022-01-15, 33.2% done, ETA December 2023
DoubleCheck=333044087,81,1 DC & TC matched, done
DoubleCheck=340830817,84,1 DC did not match suspect first test; LLTC is assigned Kriesel, 4.4% done, ETA ~Dec 20 2022
kriesel is online now   Reply With Quote
Old 2022-09-10, 15:19   #22
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

25×3×71 Posts
Default status update

These are intended to be LL DC, TC, QC as needed to provide more data on LL reliability at 100Mdigit.

DoubleCheck=332196607,81,1 assigned ll 2021-02-13, last update 2021-02-13, zero progress; kriesel completed DC, matches, done
DoubleCheck=332209313,81,1 mismatched res64s; Madpoo has TC assignment, 0.1% done
DoubleCheck=332216777,81,1 assigned ll 2020-11-14, last update 2020-11-14, zero progress; phillipsjk may start in Sep 2022
DoubleCheck=332242709,81,1 LLDC assigned Kriesel 2022-01-15, 34.7% done, ETA March 2023
DoubleCheck=332253617,81,1 LLDC assigned Kriesel 2022-01-15, 39.2% done, ETA January 2023
DoubleCheck=332341057,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress; kriesel starting a DC without assignment, ETA Sep 29 2022
DoubleCheck=332430691,81,1 assigned PRP 2021-08-22, last update 2021-08-22, zero progress
DoubleCheck=333043493,81,1 LLDC assigned Kriesel 2022-01-15, 36.2% done, ETA December 2023
DoubleCheck=340830817,84,1 DC did not match suspect first test; LLTC is assigned Kriesel, 16.7% done, ETA ~Dec 20 2022[/QUOTE]
kriesel is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Best effort: What is the priority? Aillas Operation Billion Digits 2 2010-09-30 08:38
software priority SandStar Software 10 2009-11-08 13:27
priority settings in prime95 25.9 b1? joblack Software 4 2008-12-28 22:42
Prime95 causes lag even at priority 1 ixfd64 Software 3 2006-01-23 20:57
CPU Priority (advanced menu) Software 7 2002-09-24 23:54

All times are UTC. The time now is 00:15.


Mon Sep 26 00:15:50 UTC 2022 up 38 days, 21:44, 0 users, load averages: 1.63, 1.72, 1.49

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