In the past two days, I've had GPUGrid cancel two WUs due to the fact they have already been completed. One was by a 295, the other by a 260. It sort of makes you wonder WHY they sent out two jobs at the same time.
One thing it does mean is that I am now going to check how many are required for a quorum and how often it was sent out. Oh and have set Grid up so that it only gets work WHEN I am around, to carry this out. :sneaky:
That sounds like too much effort! Is it worth it?
I have had this happen quite a bit too. I note that if you have STARTED the work unit when its mate is reported, it is still allowed to complete and credit is still awarded. They only cancel them if they have not been started when its mate is reported.
I also think they are going to run into problems with the new way they are awarding the bonus credit. I think it will lead to a lot of people aborting WUs that cannot be completed within the bonus period - either extra ones that are downloaded automatically, or aborting longer ones in favour of short ones.
In fact, from a credit point of view it is now better to let a machine run out of work for a while and then return a new WU quickly than to download several, do work constantly and return subsequent ones outside the bonus period. Giving more credit for less work is not logical and I suspect will need to change.
I've had a lot of those too and as far as I can tell, WS is right - WU already in progress will still get credit. Although I've seen some cases on their website where a seemingly completed WU got 0 credit.
One will probably see this problem more often with cards that the project does not recommend (too slow). Run GPUGrid on a card with less than 64 shaders at your own "risk".
I'm hoping my 9800GT will go quicker now, having just overclocked it by 14%.
Was: 1512000 kilohertz
Now: 1728000 kilohertz
What annoys me is that I have GPU grid set to "no new work", yet it downloads work units everytime it finishes one..... that sees me with 3 or so WU's sitting there and I am having the same thing - work units being aborted by the project..
"No New work" works here (BOINC 6.6.17).
What I can't figure out is how to have it only keep 1 WU iin the cache. Logic seems to be 1 WU per CPU core (but it doesn't run on the CPU :-\). Fortunately I only have a dual core but it must be a pain for quads.
I have only just started this so I don't know for sure, but it may work. Use the local preferences to override global settings as follows:
Set connection time to 0 (always connected)
Keep 0 days worth of cache
In theory this should mean no work is downloaded until needed.
Then set up the resource shares so that GPUGrid gets the appropriate amount. In the case of a quad with one GPU, GPUGrid is to get 20% while the projects running on the CPU get 80% between them. That's how I have mine set up and there are no debt issues (all are close to zero)
Note: this is with a 9800GT on a system with Q6600 CPU. 64 bit Linux with BOINC 6.4.5
Quote from: Webmaster Yoda on March 30, 2009, 12:57:28 AM
Set connection time to 0 (always connected)
Keep 0 days worth of cache
But doesn't this apply to all projects? - I'd like to continue to keep a couple of days cache for CPU projects.
Anyway I'll try setting the resource share lower and see if it helps.
Thanks WM
Update: set resource share to 20 (was 50) but I don't have any other CUDA work in the system at the moment (SETI CUDA app is defined but SETI is set to no new tasks). GPUGRID still downloads a second WU as soon as I allowed new tasks.
PS - is it possible to get GPUGRID added to the list of active BOINC projects on the homepage? Thanks!
Quote from: dyeman on March 30, 2009, 08:30:39 AM
But doesn't this apply to all projects? - I'd like to continue to keep a couple of days cache for CPU projects.
Yoda's talking about the option in BOINC Manager not any of the project websites. In BOINC Manager select Advanced, Preferences, Network Usage and change the connect selection there.
Indeed, if it's done at a local level, it applies to all projects you run on that particular PC. It will not affect any other PC you may have running.
Having said all that, I tried it and still ended up getting a new work unit when there was one running, so it does not quite work.
Luckily my GT9800 will do 2 work units in less than 24 hours now that it's running at 700 MHz / 1.7 GHz / 2.2GHz
OK thanks I see about the local preferences now (but as Yoda says it will also affect the CPU projects on the PC).
Oh the trials and tribulations of getting credit!