Microcontroller + GPU?

Question

Electronics

Hello. Which microcontroller to choose where is the easiest way to connect Nvidia Titan X for CUDA calculations?
Stackoverflow says that I will have to code my own drivers for my Raspberry PI (by the way, why? *nix in the same place).
PS Need for TensorFlow-gpu
PSS Yes, you need a *nix based microcontroller. And to fence your computer like the core of a robot is something at the level of "laptop with motors" ...

Answer 1

by the way, why? same *nix

Because the drivers are not only low-level and must take into account the peculiarities of the architecture, but Nvidia also has them closed.
Well, first you will need to implement the PCIE bus for your controller.
And Raspberry PI is not a microcontroller.

Answer 2

Something I have some kind of dissonance, you want to do a difficult task, a very difficult task with low-level hardware and drivers.
But at the same time, you want a ready-made neural network, TensorFlow, and at the same time on the MK, you can forget.
If you really want, that is, ready-made micros of hardware acceleration of the neural network, it is already quite possible to connect them to the MK ( https://www.terraelectronica.ru/news_utr.php ?ID=160)
Also, if you want to be compact, you can take a monster from nvidia - Jetson TX2 or TX1 - www.nvidia.ru/object/embedded-systems-dev-kits-mod...