CUDA and CUDNN Not Found in Docker Env

On AGX with custom carrier board using Jetpack 4.5 we cannot build docker image because the environment cannot find CUDA and CUDNN. Both of these have been installed successfully via SDK manager along with the runtime container.

When attempting to build a docker image, CMake has an error saying that it cannot find CUDA or CUDNN in docker environment. CUDA and CUDNN (and all SDK components) were successfully installed via SDK and we get no errors when manually installing docker dependencies. If we start our docker image that contains our environment, clone our code, run CMake, then build our program manually in the container without error.

We also tried this on a Jetson NX (4.5) and AGX with (4.6). Other than this issue, everything else appears to be fully installed with no error.

Hi @alex15, can you try setting your default docker runtime to nvidia, as shown here?

This will enable you to use CUDA/ect during docker build operations (i.e. in Dockerfiles)


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.