GPGPU CUDA or OpenCL

Hi,

I was thinking about using GPGPU in a UE project, and I was wondering if it can be done from C++ code using CUDA or OpenCL?
Also, if it does work at all, does it run the same on the shield tablet, or a nexus tablet as on a PC? (With less performance probably, and I guess CUDA could only run on tegra devices, but perhaps OpenCL can run the same on PC, and on any tablet ?)

What I mean is, can I link the libraries without problems, is it worth at all experimenting with this on PC and/or tablets, or should I just stick to pure c++ if I want to do heavy computations?

I also found this in the forums :

Is this the way to go perhaps?

Any insight would be helpful

You should choose Open CL if you want your game to run on every devices.

But CUDA gives more performances then OpenCL.
So if you’re more after an impressive result, you might want to implement your logic in both OpenCL and CUDA, to fetch the 10-50% bonus performances you can get on Nvidia cards.

Also note that new gen Nvidia cards seem to be especially better for GPGPU, since NVidia claimed a 10x performances boost for AI Computing on the Pascal architecture.