It would be a significant waste of resources to have a CPU core sitting idle while the GPU is doing its thing.
Currently, BOINC sets the priority of the CPU science tasks to "Low" (on Windows).
I wonder whether it would help if BOINC was tweaked to set the priority of the GPU task to normal or above-normal. Compile-your-own BOINC, anyone? [Edit]: It seems BOINC already does this - ca05065 says in GPU Support Forum >> Working GPU crunching times "I watched several beta GPU work units with Process Explorer and GPU-Z.
In windows 'normal' processes run with a priority of 8; Boinc CPU work units with a priority of 4 and the CPU portion of GPU work units with a priority of 6. This means that that the GPU tasks will run in preference to normal work units but after normal processes." [/Edit]

Idle? What's idle?
-----------------------------
Are you running under either Windows Vista or 7?
With these, it seems you can't run GPU work if BOINC is installed in Protected Mode (aka as a Service) ( Help Conquer Cancer >> HCC with GPU ) because services can't access the GPU under these OSs.

------------------------
I'll leave it to you guys to decide when to switch these discussions to the new GPU Support Forum

Well, I don't want to go toss out my chest and let my passions get the best of me till all the parts are in my hands BUT I can toss out a teaser:
The chips will have a W at the end of their description and the board's name ends in a letter from near the end of the alphabet and should be the first one seen in the USA and last, you know I only get all hyped up it it's a dualie!
Is that enough for now?
My daily driver, this machine, is a excellent machine.
2600K, Intel board,6 gig of excellent GSKill memory,WD Black,260GTX vid and does everything just fine but it doesn't incite passion.
The dualies get my blood running and at 60 that takes more than it used to!

From my reading, feel free to interject: If a core is reserved to support GPU crunching, the GPU jobs could run X factor faster [how much?]. This varies per science and probably for those GPU sciences rated "non-intense" it does not really matter... those doing <2% CPU load. On a 24 hour period, take how much science is done on 1 core, combining regular, randomly assigned CPU sciences with constant or intermittent streamed GPU tasks of same or varying intensity and determine if net more or less work is done.

A main question is, what continuity does WCG expect to deliver on GPU work units, daily, week, month. Also, will this be WCG only or a mix with other grids, what share as well. Personally, my impression is, that there is no one stop setting to cover it all and it has to be played by hand. Do the sums and assess if for device A, setup B will work optimally, if you wish to squeeze the max. Decide and then let it run.

For those that only want to do WCG, but do not want to have their GPU idle, set backup projects [GPUgrid e.g.] with zero project weight / resource share. If WCG has non, the client will go fetch them from your backup projects. Theory as yet not seen this in operation. Who does know is asked to share.

--//--

P.S. I've used an affinity automation tool to force multiple tasks onto one core. The results varied. Some sciences could not be bothered, and ran practically twice the ''elapsed'' time, CPU time unchanged of course. Some heavier needed a bit more ''elapsed'', but not dramatic. Kind of the hyperthreaded way without the efficiency uptick this feature delivers. Don't know how well hyperthreaded systems compare to non-hyperthreaded systems in this respect. Would these handle this better?

I still have received only one Beta task.....

Not really. When the OS needs to do something, especially with a higher priority, it will bump a wu to do this. Leaving a core free (on multicore systems) should help alleviate the bump and assist with GPU crunching.