Usually upgrading to the driver version 12.11beta4 fixes the stuck WU issue but yours is not the first case of it not working that I've heard about. What you're experiencing is actually a Windows problem and how it handles the video driver. This is a regedit fix that was put out by Microsoft to address this issue. I've applied it to my all crunchers. If you're comfortable doing a regedit here's all you need to do.
Copy and paste the ALL the text below, including the Windows Registry Editor Version 5.00 part into notepad. Rename it fix.reg
Right click on the file and open with registry editor. You'll be asked if you want to add this to the registry, select yes. Reboot and you're done. That should stop the stuck WUs.

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Watchdog]
"DisableBugCheck"="1"

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Watchdog\Display]
"EaRecovery"="0"

Thanks for fix.reg, nanoprobe.

I had to go out for a few hours and on return, found twelve more stuck for over three hours. One of them at 0% (so it didn't even start the first image).

I'm applying the fix now and will report back to you if I continue to have the same problem. Thanks again for taking the time.

For sure, all *new* tasks, downloaded from the scientists is of double image. Repair work may still circulate from before the low alert that's single image. The result log shows when there is a second image started, and visually you'd see a pause at 49.7% or so lasting a minute or more, or the GPU job switching back from GPU to CPU and onto GPU again.

Generally the jobs I see on CPU easily vary 30-40 minutes... all depends on how hard the image filters take to analyse. Shortest my quad has logged took 2.17 hours for a double and longest 3.78. The singles never touched the 2 hour mark (Linux).

Uplinger, 2 machines running GPU tasks, both had the reg fix applied and rebooted as per your instructions.

The registry fix hasn't solved the problem unfortunately.

The machines have multiple GPUs configured with app_info.xml and stuck units aren't limited to any particular GPU. stderr doesn't appear to show anything out of the ordinary and if the units are suspended then resumed, they complete within a few minutes (compared to the hours they have been sitting idle). BOINCTasks sometimes shows high CPU activity which I presume means that image one is completing.

Is there anything else I can check? Also, do you have a link for the Microsoft fix to see if there are any more pointers?

Thanks again for your time.

Hello coolstream,
I have had problems similar to those andzgrid mentions. The computer I got in 2010 always messed up the running BOINC tasks if I set the screen to power down. I had never come across this before on any computer, but after almost 2 weeks I gave up and only used manual power off for the monitor on that computer. HP had some bug in the hardware or software on that system.

Lawrence