CUDA-OPEN CL IS ENABLED - How would I know?

Question

After backing up my working OS, I attempted to try latestSG updates with Mavericks.
I know, what was I thinking...?
Mac Pro Early 2009
32 Gigs Ram
CUDA Version 5.5.28
Nvidia GTX 285
Mavericks
SGworks ok with native SG - LUTs and masks dont kill playback. Lots of caching.
Is it using open CL?
THE QUESTION: HOW WOULD I KNOW IF IT IS?
How can we check if GPU is being ultilised? Unlike all AE,PP, ME and proper adobe apps, SG doesnt give me any indication of whats going on.  This is rubbish.
If I grade same clip via Direct link plays back at one fame per second.. 
Sending a DL'ed grade back to premiere give me no image, just a 'Media Pending Link' - unless I turn off the embedded Lumetri effect.
Could anyone tell me what CUDA drivers I might roll back to to get this working?
I mean, it was tested with a specific set of 'current' drivers, right?
sidenote: After Effects CC also refuses to utilise this card, which is plenty fast for most things.

Answer 1

Same test with Mountain Lion
Mac Pro Early 2009
32 Gigs Ram
CUDA Version 5.5.28
Nvidia GTX 285
Using CUDA in PP, playback was leagues better than Open CL. But still crappy with any sort of mask in play.
Using Direct Link within Speedgrade with Single Primary layer is great.... but utterly unusuable if you add masks. (who does that eh?)
Even if you add empty grading/adjustment layers playback also nosedives, so the hope I had of isolating my masks on separate adjustment layers doesnt help, as it makes working in SG terrible.
You cant really crossfade multiple SG adjustment layers in PP, as  the vignettes need to be the same group as the grade for the transition to act it properly (or nested, but that's not the point of this workflow).
When you talk about using Nvidia Cuda GPU accelaration not being available for Speedgrade, how does impact on it's use along the whole pipeline? You need to be clear about where Cuda/CL impacts on performace, and not waste everyone's time discovering the grizzly truth. How many posts are reporting the same problems, again and again.
What if you are outputting your graded edit in Premiere to AME (which also has CUDA/CL options?  If I render SG graded edits in AME from Premiere - using Open CL adds about 50% to the rendering times. 
Despite having 32gigs of available ram, SG insists on caching to disk - why?
Admit that the performace using masks is Direct Mode is not ready for production, and stop trying to pretend it is.
Update the information links on this forum, as many are out of date, or point to previous versions.