I have a GTX 285 for crunshing the cuda gpus (no other projects running, no cpu jobs atm).
Very often workunits get supended before they are finished (happens between 1% and 99% finished).
When I have 20 wus to crunch, maybe 70% of them will get suspended before they are finished.
Any idea how I can stop this behaviour? It costs a lot of crunching time, since the suspended wus start from the beginning when its their turn again.

I have a GTX 285 for crunshing the cuda gpus (no other projects running, no cpu jobs atm).
Very often workunits get supended before they are finished (happens between 1% and 99% finished).
When I have 20 wus to crunch, maybe 70% of them will get suspended before they are finished.
Any idea how I can stop this behaviour? It costs a lot of crunching time, since the suspended wus start from the beginning when its their turn again.

You need to "leave applications in memory" in your preferences. And there is a preference to run on the GPU always in the BOINC manager.

Thanks for you answer.
I already had the cpu preference on active. I tried leaving applications in memory, the problem is that with a lot of suspended wu's I'm starting to get out of memory errors with new wus.
I'll try to keep the queue now as short as possible to avoid this, I hope that will work.

Reducing the amount of wus requested at a time helped, since the queue is now too small to cause any problems :)