Docker environment you can use to build CUDA-compatible Machine Learning tools and libraries for NVidia Jetson Nano boards.
The environment in which the libraries are built is the combination of host and container libraries. In particular,
Library | Version |
---|---|
CUDA | 10.2(.89) |
CuDNN | 8.0(.0.180) |
Check the lists dependencies-apt.txt
and dependencies-py3.txt
for the container libraries.
The Docker image (i.e., environment) can be built on any machine (i.e., any architecture).
The libraries build when the image is run, and this can only happen on a machine with arm64v8
architecture and with the proper version of CUDA and CuDNN installed.
NOTE: This Docker image DOES NOT have CUDA/CuDNN installed in it. CUDA and CuDNN are mounted
by the nvidia
runtime for Docker.
Build the environment image using the command:
dts devel build
Build a library using the command:
dts devel run -L <library_name> -- -v $(pwd)/dist:/out
where, library_name
is one of those available in the
/launchers
directory of this repository.
The final python wheel will be available in the directory
/dist
of this repository once the building has completed.
As of January 2020, it is possible to install the JetPack v4.4.1 on an NVidia Jetson AGX Xavier.