mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing > GpuOwl

Reply
 
Thread Tools
Old 2020-08-05, 15:54   #2377
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

1,499 Posts
Default

Quote:
Originally Posted by kriesel
...Then after the primality test is finished, the proof file is generated, and the proof file must be uploaded to the server, either through gpuowl's primenet.py or through the uploader program George provided, before a CERT verification can be run on it to validate the primality test.
"gpuOwl's primenet.py." Python. I do not know anything about Python. I prefer Perl. Maybe a compiled binary would be the best way for all, one for Linux and one for Windows.

Until late yesterday evening, I was not aware that I was four updates behind with gpuOwl. Now, updated. I decided to test my setup by running a PRP-CF test.

In the folder created using the exponent as a title, resides another folder titled "proof." In this folder is 154 items and growing rapidly. This must be the proof data which would be uploaded. Since this particular exponent has three known factors, perhaps the transfer may not be needed. The test itself, is running with no problems.
storm5510 is offline   Reply With Quote
Old 2020-08-05, 16:09   #2378
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

23×19×29 Posts
Default

Quote:
Originally Posted by storm5510 View Post
A person runs the PRP test, another runs the DC test, and still another will be assigned the CERT verification test. I believe I may see where this is headed, the elimination of LL and LL-DC.
Four likely cases in the short term, and a possible fifth:
1) First primality test is PRP without -proof. It requires a PRP-DC. The PRP-DC is also without -proof. Generally these will be composite and match residues and be sufficient. Cost is 2 PRPs, slightly less cost than 2 LL's plus the occasional LL TC etc which is ~2.04 on the average, but more reliable because of the GEC.

2) First primality test is PRP without -proof. It requires a PRP-DC. The PRP-DC is done with -proof. A CERT run will follow. Cost is ~2.01 PRPs. Generally the PRP's will be composite and match residues. Even more reliable than traditional PRP & PRPDC.

3) First primality test is PRP with -proof. A CERT run follows. Highly reliable, more reliable than case 1's PRP & PRPDC. Cost ~1.01 PRPs, about half that of case 1 or doing LL & LLDC. This is the preferred form of all future first primality tests.

4) First test was an LL. By current assignment rules an LLDC follows, not a PRP. Cost with normal error rate is ~2.04 primality tests. Less reliable and more cost than traditional PRP & PRPDC. (GEC in PRP is almost 100% error detection, while Jacobi check is 50% detection in some LL capable software and absent in other software.)

5) One could consider changing the assignment rules to allow and accept PRP with -proof and CERT as DC for an LL first test. Its higher reliability is an advantage.

In the rare case that any of the preceding yield probably-prime or prime not composite, from reliable software, multiple LL tests will be performed on a variety of hardware and software to confirm the new find. (CUDALucas will be treated skeptically due to known issues that can produce false positives.)

Last fiddled with by kriesel on 2020-08-05 at 16:09
kriesel is online now   Reply With Quote
Old 2020-08-05, 16:16   #2379
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

23·19·29 Posts
Default

Quote:
Originally Posted by storm5510 View Post
This must be the proof data which would be uploaded.
The collection of interim files get stored until the PRP is finished beyond exponent p, up to topk. Then the proof file is generated from them. The collection is multiple GB for a wavefront exponent and normal proof power, and temporary and not uploaded. The proof file is a fraction of a GB and gets uploaded.
kriesel is online now   Reply With Quote
Old 2020-08-05, 17:08   #2380
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2×4,297 Posts
Default

At the moment, I think that LL's that need a DC, should in general be left to machines that are not upgraded to v30, rather than do a fresh PRP.
But, the "strategic" LL-DC candidates (those that have suspect results from a user that is not a forumite) should go to PRP with a cert.
Uncwilly is offline   Reply With Quote
Old 2020-08-05, 17:41   #2381
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

221278 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
But, the "strategic" LL-DC candidates (those that have suspect results from a user that is not a forumite) should go to PRP with a cert.
However... That wouldn't give us data as to suspect machines. I beleive Aaron uses this data to further refine SDCs.
chalsall is offline   Reply With Quote
Old 2020-08-05, 18:20   #2382
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

859410 Posts
Default

The question comes about of what can be done about those suspect machines. A forumite can be contacted after their results are directly DC'ed, Anon can't be.
Uncwilly is offline   Reply With Quote
Old 2020-08-05, 18:49   #2383
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

29×41 Posts
Default

Quote:
Originally Posted by storm5510 View Post
"gpuOwl's primenet.py." Python. I do not know anything about Python. I prefer Perl. Maybe a compiled binary would be the best way for all, one for Linux and one for Windows.
If you prefer perl, it shouldn't be too hard to translate the python script to perl.

If you have a python or perl script, you can inspect it and see what it does. You can fix bugs or tweak the behavior. If you have an executable, you have to trust somebody else about what it does, and can't fix or alter the behavior yourself.

Arguably, a python script can be turned into an executable, there are tools for that, but I personally don't see the benefit of that transition.

Quote:
I decided to test my setup by running a PRP-CF test.
GpuOwl does not do PRP-CF AFAIK... !?

Quote:
In the folder created using the exponent as a title, resides another folder titled "proof." In this folder is 154 items and growing rapidly. This must be the proof data which would be uploaded. Since this particular exponent has three known factors, perhaps the transfer may not be needed. The test itself, is running with no problems.
Those are the temporary checkpoints that are used for the proof generation when the test is completed. Afterwards they are deleted and only the proof, much smaller, is kept and uploaded.
preda is offline   Reply With Quote
Old 2020-08-05, 19:54   #2384
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

23·19·29 Posts
Default

Quote:
Originally Posted by preda View Post
Arguably, a python script can be turned into an executable, there are tools for that, but I personally don't see the benefit of that transition.

GpuOwl does not do PRP-CF AFAIK... !?
Mihai,
I'm not sure any of my Windows systems have python installed. Standard Windows installations don't include a development environment for anything. It would be tedious to install python on every system capable of running gpuowl.
I have one perl development system and compile scripts to use on the rest.
Also there's one gpuowl-build system and copy the compiled file to the rest after putting it on a server.
It's very understandable if you don't want to compile your primenet.py, or haven't Windows set up on which to do it for others.
We Windows users will just have to work something out for that.

I looked at Storm5510's "PRP-CF" result earlier. It looks to me someone else had done a PRP-CF run before him, and his run was PRP not PRP-CF. The residues match because they're compatible type. And now ATH has a PRP run in progress on it also.

Last fiddled with by kriesel on 2020-08-05 at 20:04
kriesel is online now   Reply With Quote
Old 2020-08-05, 21:10   #2385
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

7·1,021 Posts
Default

You can use prime95 v30.2 to upload gpuowl proofs. Just make sure the proof files are in the same folder as prime95.exe. That's what I do (to get bandwidth limiting and time-of-day features).
Prime95 is offline   Reply With Quote
Old 2020-08-05, 23:34   #2386
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

1,499 Posts
Default

Quote:
Originally Posted by preda
GpuOwl does not do PRP-CF AFAIK... !?
Quote:
Originally Posted by kriesel
I looked at Storm5510's "PRP-CF" result earlier. It looks to me someone else had done a PRP-CF run before him, and his run was PRP not PRP-CF.
The worktodo line said "PRP" only. The result line says "PRP-3." Below is a test I pulled from Primenet for demonstration only. I unreserved it later.

Code:
PRP=xxxx,1,2,10369241,-1,99,0,"765044109502655639249"
The assignment in my account also reads "PRP." The reservation was from "Double-check tests on Mersenne cofactors." I do not know how a determination is made based on the work line as to the type.
storm5510 is offline   Reply With Quote
Old 2020-08-05, 23:42   #2387
paulunderwood
 
paulunderwood's Avatar
 
Sep 2002
Database er0rr

32×13×29 Posts
Default

Quote:
Originally Posted by storm5510 View Post
The worktodo line said "PRP" only. The result line says "PRP-3." Below is a test I pulled from Primenet for demonstration only. I unreserved it later.

Code:
PRP=xxxx,1,2,10369241,-1,99,0,"765044109502655639249"
The assignment in my account also reads "PRP." The reservation was from "Double-check tests on Mersenne cofactors." I do not know how a determination is made based on the work line as to the type.
It looks like the cofactor is given: 765044109502655639249
paulunderwood is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
mfakto: an OpenCL program for Mersenne prefactoring Bdot GPU Computing 1634 2020-09-10 21:40
GPUOWL AMD Windows OpenCL issues xx005fs GpuOwl 0 2019-07-26 21:37
Testing an expression for primality 1260 Software 17 2015-08-28 01:35
Testing Mersenne cofactors for primality? CRGreathouse Computer Science & Computational Number Theory 18 2013-06-08 19:12
Primality-testing program with multiple types of moduli (PFGW-related) Unregistered Information & Answers 4 2006-10-04 22:38

All times are UTC. The time now is 13:25.

Sat Sep 19 13:25:13 UTC 2020 up 9 days, 10:36, 1 user, load averages: 1.20, 1.20, 1.32

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.