docker-cuda-jetpack

Docker containers for CUDA with GPU acceleration, running on Nvidia Jetson.

Tested on a Jetson AnandTech thing, using Docker 19.03.06. Make sure you're on JetPack 4.5.

Use the build.sh script to build the containers. Then use one of the run-....sh scripts. You'll have to enter your root password.

There are three containers:

base: bare minimum to deploy pre-built CUDA app
runtime: extends base adding all shared libs from the CUDA toolkit. Use if you have pre-built apps using multiple CUDA libs
devel: the 'real thing' for development; extends runtime by adding the compiler, debugging tools, headers, static libs and the CUDA samples

Here's how you can use the devel container to run one of the CUDA samples:

./run-devel.sh
cp -r /usr/local/cuda/samples /tmp
cd /tmp/samples/5_Simulations/nbody
make
./nbody

For more details, see the nvidia-docker wiki (instead of nvcr.io/nvidia/l4t-base:r32.2, use cuda-jetpack:4.2.1-devel).

clearobject/docker-cuda-jetpack