4.9 suggestion?

Love the studio don't get me wrong.
Do not like the fact that it puts the textures for rendering on the GPU memory for Iray GPU rendering. I have an Nvidia GTX 970 4gb works really well.
My CPU however does not like to render in Iray, AMD FX 8 core, 4ghz, Heats it up and takes forever. So my suggestion and question for those of us who do not have mutiple GPU's is: Why can't the cpu handle the textures and feed them to the GPU instead of the GPU storing them all? Interactive between the cpu for texture processing and hand over allowing the GPU memory to focus on only the render itself. If there is enough speed between the cpu and gpu it should allow one to render a larger picture in the same amout of time. or render the same size in a shorter amount of time.
My rig has 32 gb ram mainly unsed, while my GPU has 4 gb ram heavily taxed. I am thinking the textures load on the 32 ram and cpu, leave my 4gb for cuda ops.
Think this may work? Most of us do not have multiple GPU's.
Comments
Simple answer...Iray is not capable of doing that. It has nothing to do with Studio and everything with the way Iray is written.
It's ctually not just iray, the PC infra structure cannot handle that, or it could but it would be very slow, in the future you might see this but the hardware is not capable of doing it today, the GPU memory and CPU memory are separate beasts (almost) and so far there is no fast way of doing what you want.
And to make things worse, you said it yourself "If there is enough speed between the cpu and gpu", there isn't, the GPU need blazing fast access to the textures, it takes time to send it from CPU to GPU memory and to make it worse you have a problem with GI, it will need pretty much random access to textures because it does not know what will be hit, a ray can bounce around like crazy so you would need to load, throw away, load again and so on to make it even slower, ray coherence algorithms can be used to solve that problem to some extent but it would mean you would have to cache rays instead and run into the same problem.
Look at it like this, having more furniture (textures) than you have room for in your house (GPU) is not a good idea, switching sofa from the one in the garage (CPU) takes time so once you changed to the sofa you like the movie is half finished already...