Running Cuda on Seti@home, one or two task per GPU will run without any problems until task is restarted.
To further complicate it, it will take a few restarts before problem occurs.
I have tested by pausing those tasks that seems to take much longer to process than the other similar tasks (I see it typically allready at task startup, as it will take a lot more than 1 minute to progress, opposed to about 30 seconds or less for the others), and then unpause them. They are at that time being processed much quicker. As I found this a bit strange, I continued testing, and the same with all tasks - they ran much quicker after they were paused and then unpaused. This I did until computer froze. Before the freeze the computer got less and less responsive, so it’s not an immediate hang.
I did this over and over again with just one task at a time, and every time it ended in computer freezing.
The problem ofcourse also happens when tasks are automatically put on hold, as tasks with earlier deadlines are put ahead in the chrunching queue. In this scenario all running tasks on all gpus are paused, and new ones started. Also here no problem with first or second restart, perhaps even more, but every time the computer was frozen again at some point. The log allways showed that the last thing before hang was a task reshuffling as described above.
I have 2 other fellow Seti@home chrunchers with the same problem, so it’s not specific to my hardware/software config:
P8P67 WS Revolution rev 3.0 B3 revision.
3 GTX 590 (same problem also with just 2 GTX 590)
1800 watt PSU
Windows 7 x64 SP1
GeForce/ION Driver v270.61 (same problems with v270.51)
Cuda 4.0 RC 2 (same problems with 3.0)
My host at Seti@home: http://setiathome.be…?hostid=5871814
It seems like immature drivers, but what is it and how to address it?