mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2020-01-17, 02:46   #1
xx005fs
 
"Eric"
Jan 2018
USA

199 Posts
Default Manual Assignment Feature Request

I discovered that CAT0 and CAT1 exponents aren't available for manual testing, and with the current performance of GPUs in LL and PRP, is it possible that users with known fast submission time received CAT0 and CAT1 exponents in manual assignments? If the assignments aren't completed on time and they just simply automatically expire?
xx005fs is offline   Reply With Quote
Old 2020-01-17, 07:31   #2
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

25×251 Posts
Default

Cat 0 and 1 are not open to manual assignments. This is clearly stated in the assignment rules.
https://www.mersenne.org/thresholds/
There is no plans to change this.

GPUs are much appreciated doing Trial Factoring (their utility in this type of testing is much better than doing primality testing.) All work is appreciated. Exponents tested in Cat 2 will help to clear milestones as much as those in Cat 1.
Uncwilly is offline   Reply With Quote
Old 2020-01-17, 08:22   #3
Fan Ming
 
Oct 2019

22·17 Posts
Default

Many GPUs also have excellent performance for primality testing now, for example, Tesla P100 and K80 on Google colab, and Radeon VII. They are mush faster than CPUs of most computers.
And Cat2 assignments are much further than current wavefront now. To avoid many assignment in Cat0 and Cat1 stuck in jams, it would be better to open Cat0 and Cat1 assignment for manual tesing if target user proved fast enough (much faster than currently required in assignment rules).
Just control the number of this assignment that can be assigned every time (for example, at most 1 unfinished Cat0 or Cat1 assignment for manual testing) to avoid malicious occupation.
Fan Ming is offline   Reply With Quote
Old 2020-01-17, 13:55   #4
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

1F6016 Posts
Default

There is a way to circumvent the system if you really want to do it. Set-up a copy of Prime95 to run and get assignments from PrimeNet. Once you obtain a Cat 1 or Cat 0 assignment, move that assignment over manually to your GPU.

While the GPU's are fast doing LL's. They tend to produce more work compare to CPU's by doing TF. If you compare the GHz-days/day doing LL or PRP that a CPU can produce vs its GHz-days/day that it can produce doing TF, then do the same for a GPU. The GPU will produce vastly more GHz-days/day doing the TF, compared to its LL output.
Uncwilly is offline   Reply With Quote
Old 2020-01-17, 14:18   #5
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

11·347 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
While the GPU's are fast doing LL's. They tend to produce more work compare to CPU's by doing TF. If you compare the GHz-days/day doing LL or PRP that a CPU can produce vs its GHz-days/day that it can produce doing TF, then do the same for a GPU. The GPU will produce vastly more GHz-days/day doing the TF, compared to its LL output.
Another way to look at it is if it's PRP you want to run, the life cycle cost effectiveness of a gpu doing PRP can be better than a cpu. I think a well tuned Radeon VII is currently the best buy. Apparently folks like Preda and Prime95 do too. GIMPS is the Great Internet Mersenne PRIME Search, after all. No GIMPS prime discovery was made by exhaustive TF, and won't be. It takes LL, or PRP confirmed by LL, to identify a rare prime. A Radeon VII can knock out a first-PRP test at the current wavefront in about a day.

Last fiddled with by kriesel on 2020-01-17 at 14:30
kriesel is offline   Reply With Quote
Old 2020-01-17, 14:19   #6
Fan Ming
 
Oct 2019

22×17 Posts
Default

There is an obstacle if a user doesn't have a computer that satisfy Cat0 or Cat1 assignment requirements. Even he/she does have, this computer can't stop doing some work otherwise it will not satisfy Cat0 or Cat1 assignment requirements one day in future.
And it's a waste for many GPUs to do trial-factoring. Many GPUs which are good at LL/PRP aren't excel at trial-factoring very much, for example, Tesla P100/K80, Radeon VII.
There are many GPUs that suitable for trial-factoring very well, but also some powerful GPUs at LL/PRP tests mainly.

Last fiddled with by Fan Ming on 2020-01-17 at 14:20 Reason: grammar
Fan Ming is offline   Reply With Quote
Old 2020-01-17, 14:33   #7
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×5×29×31 Posts
Default

Quote:
Originally Posted by kriesel View Post
GIMPS is the Great Internet Mersenne PRIME Search, after all. No GIMPS prime discovery was made by exhaustive TF, and won't be.
No, but...

As James shows on his "Economic cross-over analysis", it "makes sense" to TF to a certain level on a particular GPU before running a LL/PRP test ***on the same card***.

In the 9xM ranges we're currently working, even a P100 should TF to at least 75 "bits" before doing an LL. More candidates will be eliminated doing this.

But, of course, as you say, no MP will be found doing TF'ing. However, the next MP will be found faster TF'ing "optimally" first.

Lastly, because of Ben the goal of optimal has been thrown out the window. And, as always, people are welcome to do whatever kind of work they enjoy.
chalsall is offline   Reply With Quote
Old 2020-01-17, 14:44   #8
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

25×251 Posts
Default

Quote:
Originally Posted by Fan Ming View Post
There is an obstacle if a user doesn't have a computer that satisfy Cat0 or Cat1 assignment requirements.
If you start a machine using Prime95 "doing" Primality or DC assignments (assignment is obtained via primenet and credited to machine), and then move them to the GPU, you will rapidly earn status for this machine to do Cat 1 then Cat 0. (At least I believe this to be the case.)

Every machine will eventually get slow enough to fall out of Cat 0 and then 1.
Uncwilly is offline   Reply With Quote
Old 2020-01-17, 18:34   #9
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

1110111010012 Posts
Default

Quote:
Originally Posted by chalsall View Post
As James shows on his "Economic cross-over analysis", it "makes sense" to TF to a certain level on a particular GPU before running a LL/PRP test ***on the same card***.
It makes even more sense to have that TF occur on an RTX20xx which is considerably faster at TF and slower at PRP or P-1 and leave the Radeon VII or cpu do what it does best. I do something similar with my gpu model assortment routinely; distribute assignments with regard to TF/DP ratios that range from 0.7 to 1.3 on cpus and 11.5 to 22.5 on gpus (RTX is more like 40).

Yes, lots to consider, and too many variables to truly optimize.

Last fiddled with by kriesel on 2020-01-17 at 18:35
kriesel is offline   Reply With Quote
Old 2020-01-18, 02:21   #10
Fan Ming
 
Oct 2019

22·17 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
If you start a machine using Prime95 "doing" Primality or DC assignments (assignment is obtained via primenet and credited to machine), and then move them to the GPU, you will rapidly earn status for this machine to do Cat 1 then Cat 0. (At least I believe this to be the case.)
If you finish them on GPU, it will be credited to Manual Testing, and not the computer...
Quote:
Every machine will eventually get slow enough to fall out of Cat 0 and then 1.
But at least will not in a few months
Fan Ming is offline   Reply With Quote
Old 2020-01-19, 18:30   #11
xx005fs
 
"Eric"
Jan 2018
USA

199 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
If you start a machine using Prime95 "doing" Primality or DC assignments (assignment is obtained via primenet and credited to machine), and then move them to the GPU, you will rapidly earn status for this machine to do Cat 1 then Cat 0. (At least I believe this to be the case.)

Every machine will eventually get slow enough to fall out of Cat 0 and then 1.
How would I submit my results without it being credited to manual testing? Otherwise the CPUs will still be receiving CAT3 or CAT2 assignments and neglects the whole purpose.
xx005fs is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
manual result submission page feature request kriesel PrimeNet 3 2020-04-04 19:49
Feature request TheMawn PrimeNet 3 2013-06-17 02:32
v5.0 Feature Request Bent PrimeNet 2 2008-12-07 23:22
Feature request (Again) JuanTutors Software 7 2008-10-30 00:19
feature request tha PrimeNet 10 2003-12-12 15:37

All times are UTC. The time now is 10:41.

Sat May 30 10:41:05 UTC 2020 up 66 days, 8:14, 1 user, load averages: 2.23, 2.31, 2.13

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, 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.