To much much lag on my nvidia also. So will only be running gpu tasks at night while pc is not in use. During the day cpu task's will be running.
Hopefully the tech's fix the prob soon or I will have to stop all gpu task from running from WCG.

Can not even check mail account while gpu task's are running. GPU computation at WCG Worthless as a background task program on Nvidia card.

For some users this will be the case. We are drafting a clear statement as to why this is and will post it soon. For those users who experience this, we suggest that you set the option "Do work on my graphics card while computer is in use?". BOINC will automatically stop running the GPU task while you use your computer and start running a CPU task in its place. Then when you stop using your computer, the GPU task will resume.

I personally have this set on my home desktop and have a 1 minute delay before resuming the GPU task.

Because GPU processing can impact the users computer, running using your GPU is opt-in and running on your GPU while actively using your computer is also opt-in. This is very different than CPU processing. The benefits are clear however - the GPU contribution is already 1.5 times as large as the CPU contribution for Help Conquer Cancer and will massively accelerate the project.

I might have to watch TV on the Raspberry PI

David - I haven't set up a Windows Media Centre, so forgive the ignorance here.

I assume that you can watch tv, watch a dvd and/or listen to music. I'm assuming that both dvd and music are done via windows media player? Could you set up the cc_config to use the tag: <exclusive_gpu_app>important.exe</exclusive_gpu_app> so that the gpu stops while you are watching the movie?

And to further explain it a bit:
1) Graphics cards run data through kernels (tiny "programs" that does the same work on lots of data in parallel)
2) In a simplified way, the driver uploads data to the graphics card, asks it to run a kernel on the data and then gets feedback on what happened
3) It isn't currently possible for the GPU driver to pause and resume the work of a single kernel (for instance to suddenly paint stuff to the screen). This means that other programs that use the GPU need to do their work in between the HCC kernels. If the kernels are big and take a long time to process on the GPU that means the card is locked for longer periods of time and the user interface will appear more sluggish.

In the extreme case a kernel will lock up completely and never return. This will cause a driver crash, the screen will flash black for a short while and then a GPU failure report will be generated for the user when the driver recovers (this is the default behaviour at least - can be configured in regedit).
In the other end of the spectrum tiny kernels will require a lot of maintenance from the CPU because the driver will constantly need to instruct the GPU on what to do with what data etc. Sometimes this leads to less than 100% GPU utilization because the CPU or the PCI bus cannot keep up.
The perfect situation is one where the kernels are "just right", keeping the GPU busy, but are still short enough to allow the UI to refresh often enough. This perfect spot is different for each card series - and hence is pretty hard to hit.