PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Generalized Fermat Prime Search : GFN-14 Consecutive Primes Hunt Season is open! (Message 147257)
Posted 134 days ago by stream
Few notes regarding end of project.

1. Leading edge is currently advances at 3,5M-4M per day. Of course it may be effect of initial buffering, but I should give a warning that we may have only 3-4 days until server runs out of GFN-14 work (only 13M left).

2. When project ends, it'll be no GPU tasks on the server for few days (not counting rare GFN-14 resends). Please ensure that your clients are configured to crunch something else.

3. At some point, I'll need to schedule a maintenance on the server. I need to upgrade MariaDB and do few difficult reconfigurations. During maintenance, site and scheduler will be completely offline for few hours (file uploads will work). It'll happen when most of GFN-14 tasks will be returned and purged from database. Since I had to do dump and restore of some tables, it'll be safer to run maintenance when database contains minimal amount of data. The date of maintenance is currently unknown and will be posted here later.

4. The next project after GFN-14 will be another search for short GFNs. Since each of these tasks is very short, we'll use brand new version of Genefer which can test many candidates at once, fully utilizing your GPU. I hope it'll eliminate the need for app_config.xml. New version also contains Gerbicz checks to verify correctness of calculations. Sorry, no fast DC here, it's a different (although tied closely to Gerbicz checks) thing.

Since these primes will be very small, the purpose of the search is mostly to gather statistical information about their distribution. But who knows, may be they can became a starting point of different unusual sequences of primes.

You may see that new project will appear on the server, but no real tasks (except for possible few tests) will be sent there until I finish with maintenance and announce new project on the forum (new thread will be opened).
2) Message boards : Number crunching : Is turning on hyper threading better if you still assign number of tasks to only pysical cores? (Message 147200)
Posted 135 days ago by stream
All things being equal, if I assign a number of tasks only to the number of physical cores but turn on hyper threading for any kind of "overhead" would you generally expect better performance?

It depends on how OS kernel works. Not sure about better performance, but, at least, it does not harm much under Linux - Linux kernel is smart enough to correctly schedule tasks on different physical cores. You may quickly change run mode at will (using real cores only for LLR or all cores for sieving) without need to reboot and go to BIOS to turn HT on/off.

Windows is completely different story. I'm not sure about Windows 10, but early it did not distinguished between real and virtual/hyperthreaded cores, so you may end up running things on half of your real cores in hyperthreaded mode. I think that under Windows you must pin applications (or Boinc.exe itself, so started applications will inherit this setting) to correct set of "real" processor cores.
3) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 147076)
Posted 140 days ago by stream
Hi, i have a question. I have an ROUND OFF error with one SoB-Task running on CloudServer. Shoud i be worried?

A small error (just a bit higher then 0.4) can happen on some tests, when chosen FFT size is close to limits. When it's followed by "Disregard this error, ..." message, then it's nothing to worry about. I don't see this message in your log so I think it was a real hardware error. But since this is a cloud server, you cannot do anything anyway, so there is also nothing to worry about :) The test will be completed successfully, but it'll take few extra minutes because it has to be restarted from last known good state.

A round off error of 0.5 and above means hardware failure in almost all cases.

Why is LL2 switching to an bigger FFT Size after to many errors? Are to small FFT-Size cause errors?

Choosing correct FFT size is difficult . It must be as small as possible (because smaller is faster) but big enough to avoid these round off (loss of precision) errors. First, the program is trying to avoid a situation that initial FFT size was chosen too small, which can lead to infinite loop of retries. As a side effect, it decreases temperature of CPU (big FFTs are slower) - if an error was caused by overheating, it increases a chance of test to complete successfully.
4) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 147053)
Posted 141 days ago by stream
An interesting example: Mega Prime found by broken host.

http://www.primegrid.com/workunit.php?wuid=688110788

(Click on task number to view output).

As you can see, there is a message "Gerbicz check failed" in task output, along with some ROUND OFF (0.5) errors. The host is bad and unstable (overclocking, overheating).

"Gerbicz check failed" message is most dangerous here - it mean that error slip unnoticed through GWNUM. Without Gerbicz, incorrect intermediate result will be used in following stages of computation, generating wrong final result. Only LLR2 was able to detect error, recover, and give credit for this discovery to the original person.

In classic PrimeGrid DC scheme, this result will be inconclusive, and prime will be found by somebody else.
In projects without double-checking, like PrpNet or some other Boinc LLR projects, this prime will be missed forever.
5) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 146983)
Posted 143 days ago by stream
All LLR2 projects except PSP, SOB and WOO (which still have some old/classic LLR workunits unpurged) are updated to version 9.01


So once those old workunits are gone those will be on LLR2 9.01 as well?

Yes.

For PSP and WOO old workunits are expected to be purged in the end of December. SOB, unfortunately, may take a while.
6) Message boards : Generalized Fermat Prime Search : GFN-14 MEGA Prime Search starts December 05, 2020 (Message 146977)
Posted 143 days ago by stream
New prime at bmin+167796!

https://primes.utm.edu/primes/page.php?id=131496

Congratulations Pavel!
7) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 146960)
Posted 144 days ago by stream
All LLR2 projects except PSP, SOB and WOO (which still have some old/classic LLR workunits unpurged) are updated to version 9.01
8) Message boards : Generalized Fermat Prime Search : GFN-14 MEGA Prime Search starts December 05, 2020 (Message 146956)
Posted 144 days ago by stream
Thank you, there's still that obligatory second one left to find, just waiting for stream to post that it is unlikely to happen first !

Today, all candidates below bmin+5920 were returned as composites, so a prime found by PDW is an official first/lowest GFN-14 Mega Prime!
9) Message boards : Number crunching : LLR2 installed on all big LLR projects (Message 146827)
Posted 147 days ago by stream
Update of Proth projects started. DIV updated to version 9.01.

In new version, all tests iterations are protected by Gerbicz checks (early, due to fixed Gerbicz block size, few remaining iterations at the end of test, or "tail", were unprotected and double-checked on server). There are no user-visible changes (speed and behavior of the program should stay the same). But there were many changes on server side because server must now support both old and new types of tests (checkpoint files are not compatible), so please report if you've noticed something unexpected.
10) Message boards : Generalized Fermat Prime Search : GFN-14 MEGA Prime Search starts December 05, 2020 (Message 146795)
Posted 148 days ago by stream
Not sure if I find it relevant to continue until March when the prime has already been found.

It's for WUProp hours. Although finding an exotic prime (even not a first one) can be a good goal itself too.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 0.00, 0.02, 0.00
Generated 16 May 2021 | 2:49:45 UTC