mersenneforum.org  

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

Reply
 
Thread Tools
Old 2018-08-06, 07:44   #1563
SELROC
 

7·13·101 Posts
Default

Quote:
Originally Posted by GP2 View Post
Do you have a recent version of gpuOwL? I think recent versions have non-zero shift counts, but the earliest versions didn't.

Of course, I mean that it is operational right now, version 3.5
  Reply With Quote
Old 2018-08-06, 08:13   #1564
GP2
 
GP2's Avatar
 
Sep 2003

22×3×5×43 Posts
Default

Quote:
Originally Posted by SELROC View Post
If one computes the same number twice the second time is wasted.
Wait, are you yourself reporting the same exponent twice? A self-doublecheck?
GP2 is offline   Reply With Quote
Old 2018-08-06, 08:26   #1565
SELROC
 

7×29×31 Posts
Default

Quote:
Originally Posted by GP2 View Post
Wait, are you yourself reporting the same exponent twice? A self-doublecheck?

No. It happened to me to report an exponent which I had already reported, but that was by negligence as I didn't clear the results.txt file.
  Reply With Quote
Old 2018-08-06, 15:44   #1566
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

442110 Posts
Default

Quote:
Originally Posted by Prime95 View Post
The server is picky. It will not accept as a DC any result that is performed with the same software and same shift count.
OK, makes sense, that's an occasionally inconvenient but intended feature for sound reason. Although if there was an fft length difference, or other transform difference it might still be valid as a double check. (GpuOwL has offered -fft M61 integer transform and -fft DP floating transform and others, and with differing ways of handling carries.)

Some versions of software, including gpuOwL not so long ago, only offer zero shift.
Primenet exponent summaries eg https://www.mersenne.org/report_expo...exp_hi=&full=1 do not indicate what software was used.
One can guess by user, shift, etc, what software was used. But that info is only available after the other user's run is completed and reported. It was not available back in early February when my gpuOwL run to test fft transform limits with 153000031 was first launched.

Gpu assignments (and unassigned runs) are frequently done manually, not via primenet.
This is required for some software, since there's no primenet support in them and no available helper application to do the equivalent.

If primenet indicated software used on completed primality test assignments, it would be helpful.

A manual reservation would leave primenet with no way of knowing what software will be used or is being used. Someone seeking to select a test exponent and manually reserve a double check, and checking the exponent for any other pending assignments, may have no way of knowing if there will be a software match and shift count match.

Oh well. Certainly a line must be drawn somewhere, about accommodating special cases.
kriesel is online now   Reply With Quote
Old 2018-08-06, 17:13   #1567
SELROC
 

3×853 Posts
Default

Quote:
Originally Posted by kriesel View Post
OK, makes sense, that's an occasionally inconvenient but intended feature for sound reason. Although if there was an fft length difference, or other transform difference it might still be valid as a double check. (GpuOwL has offered -fft M61 integer transform and -fft DP floating transform and others, and with differing ways of handling carries.)

Some versions of software, including gpuOwL not so long ago, only offer zero shift.
Primenet exponent summaries eg https://www.mersenne.org/report_expo...exp_hi=&full=1 do not indicate what software was used.
One can guess by user, shift, etc, what software was used. But that info is only available after the other user's run is completed and reported. It was not available back in early February when my gpuOwL run to test fft transform limits with 153000031 was first launched.

Gpu assignments (and unassigned runs) are frequently done manually, not via primenet.
This is required for some software, since there's no primenet support in them and no available helper application to do the equivalent.

If primenet indicated software used on completed primality test assignments, it would be helpful.

A manual reservation would leave primenet with no way of knowing what software will be used or is being used. Someone seeking to select a test exponent and manually reserve a double check, and checking the exponent for any other pending assignments, may have no way of knowing if there will be a software match and shift count match.

Oh well. Certainly a line must be drawn somewhere, about accommodating special cases.



Absolutely. Doing manual testing with gpuOwl. I heard there are plans to include primenet automatic assignment in gpuOwl, but for now only manual work of reporting results is available. At this time gpuOwl falls under the "untrusted software" category for primenet.
  Reply With Quote
Old 2018-08-07, 01:15   #1568
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

4,421 Posts
Default

Quote:
Originally Posted by SELROC View Post
Absolutely. Doing manual testing with gpuOwl. I heard there are plans to include primenet automatic assignment in gpuOwl, but for now only manual work of reporting results is available. At this time gpuOwl falls under the "untrusted software" category for primenet.
I'm not sure what you mean by untrusted software, but if it has to do with the prime95/mprime style security module, all gpu applications fall in that untrusted category.

I've looked at the primenet API, and it seems to me well suited to cpu-based applications, and less well suited to gpu applications. When describing the processing capability, which of a system's collection of heterogenous gpus' properties should be given to the primenet server? Which of the attributes of the several kinds of memory for a given gpu?

These questions and trusted vs. untrusted arise, whether implementing integral Primenet support in any given gpu application, or in a single helper application that could connect the several popular gpu applications.

I have such a helper app in development, and am using it continuously on one of my systems, for several running gpu apps. It does periodic status checks on the apps, results gathering, and error screening. (Adding more log file parsers, worktodo parsers, and results file format support, one each per application type, was straightforward.) Currently it concentrates results from multiple gpu applications into a single newresults file that can be more quickly manually reported. It might get an interface to the manual reporting page instead of or before primenet. (But progress updates of gpu apps to primenet would be ideal.) The next thing I want to add is queued work run time estimation. That's particularly not simple for CUDAPm1 (even for exponents that don't crash it). It's also a necessary piece of knowing when to add more to the worktodo. Some apps implement worktodoadd, and some don't. Some provide date and time stamps, and some don't. One of these months...

Last fiddled with by kriesel on 2018-08-07 at 01:16
kriesel is online now   Reply With Quote
Old 2018-08-07, 02:22   #1569
SELROC
 

32·449 Posts
Default

Quote:
Originally Posted by kriesel View Post
I'm not sure what you mean by untrusted software, but if it has to do with the prime95/mprime style security module, all gpu applications fall in that untrusted category.

I've looked at the primenet API, and it seems to me well suited to cpu-based applications, and less well suited to gpu applications. When describing the processing capability, which of a system's collection of heterogenous gpus' properties should be given to the primenet server? Which of the attributes of the several kinds of memory for a given gpu?

These questions and trusted vs. untrusted arise, whether implementing integral Primenet support in any given gpu application, or in a single helper application that could connect the several popular gpu applications.

I have such a helper app in development, and am using it continuously on one of my systems, for several running gpu apps. It does periodic status checks on the apps, results gathering, and error screening. (Adding more log file parsers, worktodo parsers, and results file format support, one each per application type, was straightforward.) Currently it concentrates results from multiple gpu applications into a single newresults file that can be more quickly manually reported. It might get an interface to the manual reporting page instead of or before primenet. (But progress updates of gpu apps to primenet would be ideal.) The next thing I want to add is queued work run time estimation. That's particularly not simple for CUDAPm1 (even for exponents that don't crash it). It's also a necessary piece of knowing when to add more to the worktodo. Some apps implement worktodoadd, and some don't. Some provide date and time stamps, and some don't. One of these months...



I refer to the Status field in the "View your CPUs" page. It shows U for manual testing, this means the software for manual testing is untrusted.
  Reply With Quote
Old 2018-08-07, 03:50   #1570
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

4,421 Posts
Default

Quote:
Originally Posted by SELROC View Post
Absolutely. Doing manual testing with gpuOwl. I heard there are plans to include primenet automatic assignment in gpuOwl, but for now only manual work of reporting results is available. At this time gpuOwl falls under the "untrusted software" category for primenet.
And: if there is an existing LL result, a PRP assignment can not (in a test just now) be reserved for the same exponent. A separate algorithm (PRP vs LL), particularly run on a separate architecture (gpu vs cpu), and implemented by separate software (gpuOwL vs prime95 or CUDALucas), seems to me a stronger double check of primality than same software, same architecture, same algorithm, even ignoring the stronger error checking in the PRP.
kriesel is online now   Reply With Quote
Old 2018-08-07, 12:59   #1571
axn
 
axn's Avatar
 
Jun 2003

13·192 Posts
Default

Quote:
Originally Posted by kriesel View Post
And: if there is an existing LL result, a PRP assignment can not (in a test just now) be reserved for the same exponent. A separate algorithm (PRP vs LL), particularly run on a separate architecture (gpu vs cpu), and implemented by separate software (gpuOwL vs prime95 or CUDALucas), seems to me a stronger double check of primality than same software, same architecture, same algorithm, even ignoring the stronger error checking in the PRP.
Your argument is correct if we're verifying a prime. For verifying a composite, it is not. Because, now you don't have two results that match. Both could be wrong, and you wouldn't know it [Granted PRP w/ GEC is strong enough all on its own, but the principle still stands]
axn is offline   Reply With Quote
Old 2018-08-07, 13:09   #1572
SELROC
 

22·7·251 Posts
Default

Quote:
Originally Posted by SELROC View Post
I refer to the Status field in the "View your CPUs" page. It shows U for manual testing, this means the software for manual testing is untrusted.

And manual testing does not show which software is being used. This, can be very easy to resolve with a drop-down menu list of allowed software for the user to select each time an exponent is assigned for manual testing.


The history of all software used for an exponent (if the user changes software in mid computation) should be written in the results file or in checkpoint files.
  Reply With Quote
Old 2018-08-09, 05:39   #1573
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

4,421 Posts
Default

Quote:
Originally Posted by SELROC View Post
The history of all software used for an exponent (if the user changes software in mid computation) should be written in the results file or in checkpoint files.
That is included in the Mersenne Neutral Exchange format draft standard.
kriesel is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Official "Faits erronés dans de belles-lettres" thread ewmayer Lounge 39 2015-05-19 01:08
Official "all-Greek-to-me Fiction Literature and Cinema" Thread ewmayer Science & Technology 41 2014-04-16 11:54
Official "Lasciate ogne speranza" whinge-thread cheesehead Soap Box 56 2013-06-29 01:42
Official "Ernst is a deceiving bully and George is a meanie" thread cheesehead Soap Box 61 2013-06-11 04:30
Official "String copy Statement Considered Harmful" thread Dubslow Programming 19 2012-05-31 17:49

All times are UTC. The time now is 06:32.

Wed Sep 23 06:32:26 UTC 2020 up 13 days, 3:43, 0 users, load averages: 1.95, 1.97, 1.83

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.