We have recently purchased an Nvidia DGX Spark for some simulation work with Sionna and eventually Aerial integration. We have been running into software version issues between CUDA/Tensorflow/Mitsuba/Sionna. We found some discussion on sionna-rk tutorials, however there was no explicit docker image for simulation. We have had some success combining different custom versions, although we are unable to use graph compiled code (specifically decorated tf.functions). It seems to be an issue of versioning between cuda, tensorflow, and ptxas that does not arise on our existing systems.
We were wondering if there is a recommended docker or setup process for the DGX Spark? I believe there have been a few demonstrations using Sionna with the Spark so I assume there is a workflow that resolves the issues.