Cuda CPU vs GPU?

Not sure where to place this topic,

But perhaps someone could help me,
I’m running 3xGTX 1060s,
1x at a 16x slot 3.0.
1x at a 1x 3.0 slot with riser
1x at a 1x 2.0 slot with riser.

My CPU is a 1.9Ghz Xeon with 10 cores.
All GPUs are ran at 2Ghz core speed.

I’m mainly folding, but am also interested in Bitcoin mining results of this setup.
I’m not gaming.

My primary monitor is connected to the graphics card (GPU 0) inserted in the 16x slot.

Nvidia drivers and configuration allows me to choose which card I want to use for CUDA.
Either one of the 3, or CPU.

Doing folding, the pcie 1x bandwidth is often maximized, so latencies and package interference most likely will make me want to choose between the 16x slot graphics card, or the CPU.
I have Currently over 2/3rd of CPU resources unused.

What would net me the fastest, or best results?
Setting cuda to GPU 0 (16x slot), or setting it to CPU?