I moved a GTX 550TI from one computer to another. It picked up and completed a long run task but has failed on every one since.

There were zero changes to the machine until a few minutes ago when I installed the nvidia 301.42 drivers.

Here is the relevant portion of the log after a reboot and boinc "Update"

6/4/2012 8:34:38 PM | GPUGRID | Sending scheduler request: Requested by user.
6/4/2012 8:34:38 PM | GPUGRID | Requesting new tasks for NVIDIA
6/4/2012 8:34:40 PM | GPUGRID | Scheduler request completed: got 0 new tasks
6/4/2012 8:34:40 PM | GPUGRID | No tasks sent
6/4/2012 8:34:40 PM | GPUGRID | This computer has finished a daily quota of 7 tasks
6/4/2012 8:38:09 PM | GPUGRID | update requested by user
6/4/2012 8:38:14 PM | GPUGRID | Sending scheduler request: Requested by user.
6/4/2012 8:38:14 PM | GPUGRID | Requesting new tasks for NVIDIA
6/4/2012 8:38:16 PM | GPUGRID | Scheduler request completed: got 0 new tasks
6/4/2012 8:38:16 PM | GPUGRID | No tasks sent
6/4/2012 8:38:16 PM | GPUGRID | No tasks are available for ACEMD2: GPU molecular dynamics
6/4/2012 8:38:16 PM | GPUGRID | No tasks are available for ACEMD beta version
6/4/2012 8:38:16 PM | GPUGRID | This computer has finished a daily quota of 7 tasks

Disallow beta tasks for starters. They're broken. Everyone is failing those, as well as the test applications tick box.

Now, I don't know why you're failing the others, someone with more knowledge should take a look. It has the failed to enumerate device error 3.

Someone with more knowledge will post, give it some time.

Cheers

I just looked at the listing of my recent workunits.

ACEMD beta version v6.45 (cuda40) always gave Error while computing, on both of my desktop computers.

ACEMD beta version v6.43 (cuda42) also only gave errors, but only one of the computers ran any.

Long runs (8-12 hours on fastest card) v6.16 (cuda31) usually gave Completed and validated; only one failure.

Are the most recent applications unable to run correctly on a GTS 450 or a GT 440?

At least all the failed runs took less than 30 seconds each.

Using a 301.42 driver on one computer, but just noticed that the other is still using 285.62.

Then you may want to add a new option - allow only as many beta tasks as will not interfere with getting a normal flow of non-beta tasks.

That way, users could still help with beta testing, without taking much time away from the non-beta tasks they do.

Something I've seen on other BOINC projects that you might consider - if a beta test workunit fails, don't count that against the reliability of that host; especially if all other hosts that try copies of that workunit fail also.