Hello. Posting here as I'm having an issue with my newest addition to my BOINC systems. This week I finished building a "budget" AMD 945 with HD 5870 GPU, running Windows XP.

The problem I'm having is that most of the Moo! GPU work units are stopping around 98%. Yesterday I had 20+ units sitting between 95-98% while the system was working on something else. At 25 minutes worked each, that is more than 8 hours of work not yet counted for anything. All of the Moo! units are showing up as High Priority. Eventually, it will start finishing the last 2-3% of the unit and then start a new one. It seems to me that there is a timer of some sort causing this, as ALL of the "waiting" work have stopped at an exact minute of time worked (IE 25:00 minutes, 19:00 minutes).

Generally speaking, it's not much of an issue as the work will complete, but in the event of a system failure, there is a potential at 8-10 or more hours of completely lost work.

Any ideas as to why this is happening? Is there any way to not have all the new units be marked with High Priority so it completes the others first?

Thanks for any help you can provide![/img]

Hi,

Hmm, I think I have sometimes seen this kind of behaviour. It might be a sort of bug associated with High Priority detection or handling logic in BOINC Client where it prematurely switches tasks.

All this is controlled by BOINC Client so unfortunately our apps can't do anything to trigger or prevent this. Only thing that might affect things is our server settings and buffer levels we allow.

Maybe you should try to lower your cache levels in BOINC Client? This would hopefully limit the HP triggering and thus avoid these problems.

-w

I also received a response from one of the people marked as "Support" on the BOINC page. He was very helpful and helped steer me in this direction (cache issue). Turns out that I had set the cache to 4 days and that was causing the problem. If I suspended all but about 9 hours of work, the Moo tasks ran without issue. I then changed my cache setting to 1 day and the client remains working as intended.

Thanks for your response :)