Well, all of it anyway. See attached for proof: note that the manager thinks it's paused, but one of the threads is still running. This continues indefinitely unless I stop the client service.

http://bit.ly/Wi0e4r

bick-ubtu3:~$ cat /etc/lsb-release 
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=12.04
DISTRIB_CODENAME=precise
DISTRIB_DESCRIPTION="Ubuntu 12.04.1 LTS"
bick-ubtu3:~$ aptitude show boinc-client
Package: boinc-client                    
State: installed
Automatically installed: yes
Version: 7.0.44+dfsg-0~989~precise1

All that I can see in the minute info you've given, is that it's the science application that doesn't stop.

So first things first:
- Your preferences state that the GPU runs based on preferences?
- You installed Berkeley BOINC?
- This apparently being a beta app running on the GPU, you reported this behaviour at the Milkyway project first? If not, please do so.

It's very much possible that the application does not follow the boinc_exit() command that the client is sending it.

---

I just want to say that happens to me too. Exactly the same.
What can we do to fix it?

I fixed it, it is easy. I post it here for other users:

My computer has 8 CPUs and 1 NVIDIA card.
When the error happened BOINC was configured as:
-- Run based on preferences.
-- Use GPU based on preferences.
-- Don't use GPU while computer is in use
-- Computing allowed while processor usage is less than 15 percent.

When BOINC started crunching, somehow, the Milkyway GPU task was taking (together with other regular computer processes) more than 15% of the global CPU usage. Then BOINC stopped all the tasks (CPU busy) but the Milkyway task somehow continued working indefinitely, until I closed the client or interrupted the task manually.
Raising the processor usage limit to 25% the problem disappear.