mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-07-29, 18:36   #177
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

7×1,021 Posts
Default

Quote:
Originally Posted by chalsall View Post
Hey George. More data...

So, my Winblows machine has done two CERT runs in the last day. 97265881 matched, while 97282747 didn't (your run confirmed the original PRP run was good).

Neither of these went through the Proxy. Direct comms with Primenet.

Is it possible this new code path is exercising different part(s) of my kit, and I might need to service it? I /think/its history with DC'ing has been flawless.
CERT runs the exact same FFT code as DC'ing except the exponents are a bit larger. I need to get v30.2 released which properly MD5 checks the downloaded starting value.
Prime95 is online now   Reply With Quote
Old 2020-07-29, 18:42   #178
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

7·1,021 Posts
Default

Quote:
Originally Posted by kriesel View Post
CERTs are credited as PRP-DC or as PRP-cofactor-DC. I didn't see the need for yet another stats type.

Quote:
There is currently no manual CERT assignment, and that may be by design. https://www.mersenne.org/manual_assignment/.
I don't think CERTs will ever be available as manual assignments. CERTs are quick turnaround items. Manual users are often not quick.

Now gpuowl with python scripts feeding it is more like a prime95 setup than a manual user. I could see doing CERTs with gpuowl.
Prime95 is online now   Reply With Quote
Old 2020-07-29, 18:52   #179
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

22·97 Posts
Default A sea change.

It occurred to me yesterday that there may soon come a point where people start thinking about discontinuing LL work to versions of prime95 < 30 because it would be almost as much work to DC their output as it is to check a new number.

Maybe they would be forced to do only doublechecks since they would basically be the only ones producing residues to DC in the first place.
Aramis Wyler is offline   Reply With Quote
Old 2020-07-29, 20:26   #180
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2·4,297 Posts
Default

And does the assignment code also prevent users from getting certs for their own PRP runs?

And as mentioned/noted previously all PRP's that are being handed out to V30 machines get a VDF generated, even if the PRP is a DC. Will these eventually get the certs run?
Uncwilly is online now   Reply With Quote
Old 2020-07-29, 22:42   #181
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

7·1,021 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
And does the assignment code also prevent users from getting certs for their own PRP runs?
No. You cannot game the system even knowing the full contents of the proof file.

Quote:
And as mentioned/noted previously all PRP's that are being handed out to V30 machines get a VDF generated, even if the PRP is a DC. Will these eventually get the certs run?
Yes, ought to happen now though I did not test this.
Prime95 is online now   Reply With Quote
Old 2020-07-30, 00:02   #182
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

100101001012 Posts
Default

Quote:
Originally Posted by Prime95 View Post
I could see doing CERTs with gpuowl.
Yes, GpuOwl processing CERTs would be possible, but is that necessary or useful? I think the whole supply of CERT work that we can expect in the future can be easily handled by mprime.

Also, CERTs being short tasks, they could be an ideal work type for new users that have the bad habit of being ephemeral. (CERTs could also be used for initial validation of new hardware instead of DCs). This way the amount of work lost because "user left half-way through the test" would be much lower.

And I expect new users to be more likely mprime users than GpuOwl users.

And now we see that one heavy-weight GpuOwl deployment doing CERTs would "destroy" the valuable limited supply of CERTs at an alarming rate :)

Last fiddled with by preda on 2020-07-30 at 00:06
preda is offline   Reply With Quote
Old 2020-07-30, 00:20   #183
Runtime Error
 
Sep 2017
USA

11×13 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Quote:
Originally Posted by Uncwilly View Post
And as mentioned/noted previously all PRP's that are being handed out to V30 machines get a VDF generated, even if the PRP is a DC. Will these eventually get the certs run?
Yes, ought to happen now though I did not test this.
Just in case someone is looking for an example, I recently turned in 87829211 along with a proof file. The cert run hasn't been assigned yet. But the couple first time checks w/ proofs that I uploaded at the same time have all been run already and matched.

Last fiddled with by Runtime Error on 2020-07-30 at 00:24 Reason: a word
Runtime Error is offline   Reply With Quote
Old 2020-07-30, 00:22   #184
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

118910 Posts
Default

Quote:
Originally Posted by Prime95 View Post
CERT runs the exact same FFT code as DC'ing except the exponents are a bit larger. I need to get v30.2 released which properly MD5 checks the downloaded starting value.
CERTs could (and should?) use the same GEC as the normal PRP iterations. This way the correctness of the CERT work can be verified locally (that's what GEC does) before sending the CERT result.

Assuming the proof is good, we then have only three possibilities:
1. data corrupted during transfer -- detected through MD5 checksums
2. the CERT iterations have errors -- detected by GEC and fixed through rollback
3. something else is wrong (server-side/bug?)

Last fiddled with by preda on 2020-07-30 at 00:26
preda is offline   Reply With Quote
Old 2020-07-30, 00:52   #185
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

7×1,021 Posts
Default

Quote:
Originally Posted by preda View Post
CERTs could (and should?) use the same GEC as the normal PRP iterations. This way the correctness of the CERT work can be verified locally (that's what GEC does) before sending the CERT result.

Assuming the proof is good, we then have only three possibilities:
1. data corrupted during transfer -- detected through MD5 checksums
2. the CERT iterations have errors -- detected by GEC and fixed through rollback
3. something else is wrong (server-side/bug?)
I did not implement GEC for CERT work. I did not think it was worth the bother. What I did implement was shift count. If one CERT fails another is run with a different shift count.
Prime95 is online now   Reply With Quote
Old 2020-07-30, 01:39   #186
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

157538 Posts
Default

Quote:
Originally Posted by Runtime Error View Post
Just in case someone is looking for an example, I recently turned in 87829211 along with a proof file. The cert run hasn't been assigned yet. But the couple first time checks w/ proofs that I uploaded at the same time have all been run already and matched.
I'm attempting a fix now.
Prime95 is online now   Reply With Quote
Old 2020-07-30, 04:22   #187
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

104708 Posts
Default

Quote:
Originally Posted by Prime95 View Post
I did not implement GEC for CERT work. I did not think it was worth the bother. What I did implement was shift count. If one CERT fails another is run with a different shift count.
Some Certs will land on old hardware that may be less reliable, or proof files come from slow mini hardware after Mlucas adopts it; Odroid, RPi, smartphones, possibly at low proof power values, for which the verification will take longer.

Last fiddled with by kriesel on 2020-07-30 at 04:26
kriesel is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Your help wanted - Let's buy GIMPS a KNL development system! airsquirrels Hardware 313 2019-10-29 22:51
Is GMP-ECM still under active development? mathwiz GMP-ECM 0 2019-05-15 01:06
LLR 3.8.6 Development version Jean Penné Software 0 2011-06-16 20:05
LLR 3.8.5 Development version Jean Penné Software 6 2011-04-28 06:21
LLR 3.8.4 development version is available! Jean Penné Software 4 2010-11-14 17:32

All times are UTC. The time now is 20:00.

Sat Sep 19 20:00:04 UTC 2020 up 9 days, 17:11, 1 user, load averages: 1.23, 1.44, 1.56

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.