Docker#

Two Compose configurations are provided: a CPU preview image and a GPU variant with the Sionna RT extra.

CPU preview image#

docker compose up --build
curl http://127.0.0.1:8765/api/health

Open http://127.0.0.1:8765 and stop the service with docker compose down. Compose reads supported OWRT settings from the shell or an ignored .env file. OWRT_PORT changes the published host port and OWRT_BIND_ADDRESS changes the host interface. The application listens on all interfaces only inside the container; by default the host publishes it on loopback.

Sionna RT variant#

docker compose -f compose.yaml -f compose.gpu.yaml up --build
curl http://127.0.0.1:8765/api/health

The GPU variant installs the Python rt extra and requests all NVIDIA GPUs. It does not install CUDA or an NVIDIA driver. GPU execution requires a compatible host driver, Docker Engine and the NVIDIA Container Toolkit. /api/health reports whether the Sionna runtime is installed, and each completed Sionna simulation records the selected Mitsuba CPU or CUDA variant in its metadata.

What the images contain#

Both images run as the unprivileged owrt user and copy only the package source, so source data, caches and generated artifacts stay out of them. Compiled scenes are cached inside the container under /app/.cache, and batch outputs persist in the owrt-datasets Docker volume.