Question regarding TK-1 memory split

Hi there,

I’m about to port a working application to a new platform in need of more computing cores to speed it up, tk-1 is a promising contender.
Unfortunately I haven’t found the information that I need about the memory allocation by the host cpu and the device gpu on the nvidia website. (I’m not a registered developer, yet)

So, the tk-1 has 2 GB of ram, how much ram can be assigned at the max to the device gpu? Is all device memory automatically shared memory which is also accessible by the host cpu?
What is the default memory split (host, device, shared) in numbers? Can I freely change it?

I do not need any video output at all, can I use the graphic memory for program data instead or is a certain amount of memory always reserved for video?

Thank you very much in advance!


A partial answer is that Jetson uses a single 2GiB chunk of RAM which is consumed by the sum total of operating system, applications, graphics, and GPU. There is no dedicated RAM. However much RAM the operating system uses implies how much of the 2GiB of RAM is left. Cut down the operating system or applications running on it and more becomes available for everything else, including GPU. How much is that? I don’t know…perhaps someone who has X11 removed or disabled could give an approximate amount of RAM available.

Graphics is running under X11, but removing X11 should not be too difficult…console works fine, as well as serial console or ssh. I suppose there is a small amount of RAM used for the regular console, but it wouldn’t be much.