Author |
Message |
|
I am seeing work units that are taking 25-30 hours to complete. At first it was my GTX 260s but now I see my GTX 460s getting them. I can not find a mention anywhere in the forums. One of my team mates also got one that took 31 hours. Just wanting to know if these 30 hour work units are intentional.
Cheers!
____________
|
|
|
|
Hello: I looked at their tasks with the GTX460 and do not understand it takes more than I do with my GTX295.
Something is wrong with this GTX460 configured not perform properly. Greetings. |
|
|
|
Hello!
I am starting back up after taking the summer off due to 100f degree weather. The GTX 460s have a history here, 9m and 13m points. I do not remember work units ever being over 18-20 hours on them. So wondering if something has changed.
Thanks for reply.
Cheers!
____________
|
|
|
MarkJ Volunteer moderator Volunteer tester Send message
Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level
Scientific publications
|
Hello!
I am starting back up after taking the summer off due to 100f degree weather. The GTX 460s have a history here, 9m and 13m points. I do not remember work units ever being over 18-20 hours on them. So wondering if something has changed.
Thanks for reply.
Cheers!
There are now "long" work units designed for the high-end cards.
Log in to the website, go to your account and select GPUgrid preferences and then untick the box for "long" work units. Click save and tell your BOINC Client to update to disable them.
____________
BOINC blog |
|
|
|
I am seeing work units that are taking 25-30 hours to complete. At first it was my GTX 260s but now I see my GTX 460s getting them. I can not find a mention anywhere in the forums. One of my team mates also got one that took 31 hours. Just wanting to know if these 30 hour work units are intentional.
Cheers!
Looks OK seeing that you are running at near stock speed. Your not getting your full bonus because you are holding unstarted WU's too long.
In BOINC Set Connect to 0
Additional Work Buffer 0
This will make sure you have completed one WU before downloading another one.
I also have mine clocked to:
GPU 850
Memory Stock (used to have it at 2025)
Shaders 1700 they are linked to GPU anyway.
Use EVGA Precision
Oh and then join my team, won't make your crunching faster but you can enjoy the music. :)
____________
Radio Caroline, the world's most famous offshore pirate radio station.
Great music since April 1964. Support Radio Caroline Team -
Radio Caroline |
|
|
|
Thanks for the reply's and the team invite!
This is a new Linux build and so far I have not gotten Coolbits working for overclocking because I have had other issues to deal with. One being whether the work units I am getting are really supposed to take 30+ hours or if something is wrong with my machine. I am familiar with the newer Long Work units but in the past and some even now have taken 20-24 hours max. I now have a WU that has run 36 hours and is 62% done. Everything I can see in Linux says the card is NOT down clocking.
I changed my settings for the cache or queue here on web site but it is still sending 2 WUs. I do not want to change it on Boinc because I want to keep a 2-3 day cache for WCG which is my main project.
Thanks for the help. I may either go back to the shorter WUs on the 260s or move them to E@H.
Cheers!!
____________
|
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
When I had my GTX260 on Ubuntu 11.04 (a few weeks ago), some long tasks were also taking around that time - 25 to 30h. Even with a zero cache that would miss the 24h 50% credit bonus, but they did get the 25% bonus. Running long tasks and keeping a large cache would mean you could never get such a bonus so you would be as well sticking to the short tasks.
Unfortunately you cannot set cache on a project level. Local cache levels (Boinc Manager) override web settings.
I still recommend setting a low cache for all projects; neither GPUGrid or the WCG have regular or sustained downtime. There really is no reason for keeping a large cache. I hardly ever keep a cache of over 0.5days (mostly 0.01 to 0.2days) and I crunch for WCG as well. In fact having a large cache at WCG will prevent you getting Beta tasks there, and interfere with the amount of CEP2 tasks you run (default is one in cache, not one running).
Keeping a task in cache and not running it for 2days is not good for you and might be detrimental to the project; despite the now 5day return time GPUGrid needs tasks to be returned ASAP.
If you really feel you must have a cache of WCG tasks can I suggest you manually control it by suspending GPUGrid tasks, increasing the cache, download a cache of WCG tasks, then set a low cache (0.1days) and enable GPUGrid tasks again. If you do this twice a week you should keep a large supply of WCG tasks and still be able to have one GPUGrid task running without another in cache.
GPUGrid on Linux still uses the 6.14app, so it is still recommended to free a CPU core and use SWAN_SYNC if you use Linux and have a Fermi. If you crunch CPU tasks and don't do this (or alternatively write a script to automatically set the nice value for each task) then tasks will take significantly longer than they should do. I'm not exactly sure of the importance for a GTX260 compared to a Fermi card, but I expect it may still be significant, and even a 5% improvement might have seen a 25h tasks finish inside a day.
Just running short tasks should allow you to complete most tasks within 12h on your GTX260, but the cache could still come into play if you are using all 6CPU cores to the max; this would slow any GPUGrid tasks down. So recommended settings should still be used. Not all short tasks are the same length (GPUGrid does a variety of different research).
You seem to be pulling cards in and out or swapping operating systems, and I now see you aborting tasks:
Aborted by user 175,577.88 - after 48h? |
|
|
|
Thanks for the advise.
____________
|
|
|