CUDA (GPU access) for mainline kernel


I am running the mainline kernel on the TK1 board and so far everyhting runs as aspected. Now i want to access the GPU for some math operations. It seems that the only way to do so is using CUDA which requires the nvidia kernel module and this module seems to be only available for kernel 3.10 (L4T)?
Is there any way for me to profit from the GPU with my current setup?

I think you will only get hardware GPU access via the video driver, and the driver is tied to the X11 ABI (it’s a drop in replacement for the Nouveau frame buffer driver). So technically I suspect it is Xorg which holds back using that driver (which is probably tied to the kernel version, and backing out ABI version would be a very invasive task).

Do i really need X11? My board doesnt have a display connected.

Realize that X11 is a framework for accessing video frame buffers, supplemented either by a software device emulation or an actual hardware driver. Admittedly, X11 is overkill for just pure GPU access to do CUDA, but you don’t have to actually install any of the components for display (e.g., no desktop manager or login manager is required). So far as I know CUDA must always match the nVidia video driver release, and the video driver release must match the ABI of the Xorg X11 server. Display does not necessarily need to be configured.