Advanced search

Message boards : Number crunching : Why different credits for same "long runs" ?

Author Message
Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,985,645,176
RAC: 30,571,697
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 40260 - Posted: 25 Feb 2015 | 19:10:42 UTC

just out of curiosity, I would like to know why for the first few long runs [(Long runs (8-12 hours on fastest card) v8.47 (cuda65)] I ran on my computer I received 162.500 points each, and for all following ones 135.000 points. They all ran about same time, so they definitely were not that different to each other like the difference in the credit points would suggest.

[CSF] Thomas H.V. DUPONT
Send message
Joined: 20 Jul 14
Posts: 732
Credit: 115,270,366
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 40261 - Posted: 25 Feb 2015 | 19:23:09 UTC - in response to Message 40260.
Last modified: 25 Feb 2015 | 19:24:47 UTC

It depends on the time you sending back your WU's.
Bonuses are credited when you return your WU's fast.
Taking your example, you were credited with a 20% bonus for WU's that you have returned within 48 hours.
If I don't talk nonsense, but the GPUG staff should complete (or correct) my reply :)
____________
[CSF] Thomas H.V. Dupont
Founder of the team CRUNCHERS SANS FRONTIERES 2.0
www.crunchersansfrontieres

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,985,645,176
RAC: 30,571,697
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 40262 - Posted: 25 Feb 2015 | 20:20:56 UTC - in response to Message 40261.

hm, I see.

Since computation time for all the tasks was about the same (some 35 hrs. with my Nvidia Quadro K620)), and the computation was virtually not interrupted, the only reason for the "delay" beyond 48 hrs. must be that some of the new tasks were downloaded long time before the current one got finished.
This is, in fact, what I had noticed lately: short time after the computation of a given job has started, the next one was downloaded and remained in "ready to start" status for quite some time, i.e. until the previous one was completed.
However, if the overall time for the bonus starts counting at the moment the task is downloaded, but cannot start yet because another task is still at work, then of course the 48 hours would be exceeded in most of the cases.
Hence, it would be more fair if such new jobs are downloaded once the previous one has been uploaded.
I guess that most of the participants have only 1 GPU in their computer, so only one job can be worked on after the other.

neilp62
Send message
Joined: 23 Nov 10
Posts: 14
Credit: 7,876,790,536
RAC: 1
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 40263 - Posted: 25 Feb 2015 | 20:49:51 UTC - in response to Message 40262.

Erich, try setting your Boinc minimum work buffer to a small fraction of a 24-hour day (e.g., I use 0.05 for 72 minutes). In my case, the Boinc client will not download the next long run task until there is only 70 to 90 minutes of computation left on the active task. This works well regardless of the number of GPUs present. Just make certain to choose a buffer period that is longer than the time required to download a long run task, so that your GPU is not unnecessarily idle.

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,985,645,176
RAC: 30,571,697
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 40266 - Posted: 26 Feb 2015 | 5:36:58 UTC - in response to Message 40263.

Erich, try setting your Boinc minimum work buffer to a small fraction of a 24-hour day (e.g., I use 0.05 for 72 minutes). ...

thanks for the hint. I just made this adaption now, so let's wait and see what happens :-)

[CSF] Thomas H.V. DUPONT
Send message
Joined: 20 Jul 14
Posts: 732
Credit: 115,270,366
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 40267 - Posted: 26 Feb 2015 | 6:32:40 UTC

For me, it's 0.01 day :p (30mns max before the end of a WU)
____________
[CSF] Thomas H.V. Dupont
Founder of the team CRUNCHERS SANS FRONTIERES 2.0
www.crunchersansfrontieres

Jacob Klein
Send message
Joined: 11 Oct 08
Posts: 1127
Credit: 1,901,927,545
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 40270 - Posted: 27 Feb 2015 | 3:46:46 UTC
Last modified: 27 Feb 2015 | 3:48:18 UTC

Note: 0.01 day = 14.4 mins

Also, for the original poster, read the FAQs forum, especially this page that talks about Credits, and more importantly, Deadlines (and bonuses):
http://www.gpugrid.net/forum_thread.php?id=2572

[CSF] Thomas H.V. DUPONT
Send message
Joined: 20 Jul 14
Posts: 732
Credit: 115,270,366
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 40271 - Posted: 27 Feb 2015 | 6:40:15 UTC - in response to Message 40270.

Note: 0.01 day = 14.4 mins

Also, for the original poster, read the FAQs forum, especially this page that talks about Credits, and more importantly, Deadlines (and bonuses):
http://www.gpugrid.net/forum_thread.php?id=2572

Thanks so much Jacob for the link :)
____________
[CSF] Thomas H.V. Dupont
Founder of the team CRUNCHERS SANS FRONTIERES 2.0
www.crunchersansfrontieres

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,985,645,176
RAC: 30,571,697
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 40272 - Posted: 27 Feb 2015 | 8:42:44 UTC - in response to Message 40270.

Note: 0.01 day = 14.4 mins

Also, for the original poster, read the FAQs forum, especially this page that talks about Credits, and more importantly, Deadlines (and bonuses):
http://www.gpugrid.net/forum_thread.php?id=2572

many thanks for the clarification; sorry that I did not catch this earlier.
Too bad that I have only a slow GPU (Nvidia Quadro K620), so a "long run" task, without any break, takes around 36 hrs. But at least I should get the 48-hrs-bonus of +25% :-)

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,985,645,176
RAC: 30,571,697
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 40472 - Posted: 15 Mar 2015 | 8:51:00 UTC

hm, an unclearness came up with a recent "long runs" from Noelia: I finished "e10s4_e1s15f26-NOELIA_31x4-1-2-RND0004_1" within some 26 hours. In case of previous "long runs" by Gerard, for those I got 162.500 points credit, i.e. 131.250 + 25% for finishing within 48 hrs. For the task mentioned above, I got only 131.250, no extra bonus 25%. Why so?

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 40479 - Posted: 15 Mar 2015 | 19:23:44 UTC - in response to Message 40472.
Last modified: 15 Mar 2015 | 19:25:14 UTC

hm, an unclearness came up with a recent "long runs" from Noelia: I finished "e10s4_e1s15f26-NOELIA_31x4-1-2-RND0004_1" within some 26 hours. In case of previous "long runs" by Gerard, for those I got 162.500 points credit, i.e. 131.250 + 25% for finishing within 48 hrs. For the task mentioned above, I got only 131.250, no extra bonus 25%. Why so?

You did receive the 25% bonus credit actually, but the base credits for this batch are 20% less than the Gerard's you've referenced.

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,985,645,176
RAC: 30,571,697
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 40480 - Posted: 15 Mar 2015 | 19:37:51 UTC - in response to Message 40479.
Last modified: 15 Mar 2015 | 19:38:37 UTC

You did receive the 25% bonus credit actually, but the base credits for this batch are 20% less than the Gerard's you've referenced.


oh aha, this makes it clear. Thanks for the clarification :-)

Post to thread

Message boards : Number crunching : Why different credits for same "long runs" ?

//