Which base image should I use for a custom pip install?
Isaac Sim 5.0.0 pip wheels are built for Python 3.11 and GLIBC 2.35+
(manylinux_2_35_x86_64). Ubuntu 20.04 ships GLIBC 2.31, which is too old –
pip will not find any matching wheels. Ubuntu 22.04 ships GLIBC 2.35 and works.
A good starting point:
FROM nvcr.io/nvidia/cuda:12.6.0-base-ubuntu22.04
ENV NVIDIA_VISIBLE_DEVICES=all
ENV NVIDIA_DRIVER_CAPABILITIES=all
RUN apt-get update && apt-get install -y --no-install-recommends \
python3.11 python3.11-venv python3-pip \
libatomic1 libegl1 libgl1 libglu1-mesa libglx0 libgomp1 libsm6 \
libxi6 libxrandr2 libxt6 unzip ca-certificates libglib2.0-0 \
&& rm -rf /var/lib/apt/lists/*
RUN python3.11 -m venv /opt/isaacsim-env
RUN /opt/isaacsim-env/bin/pip install --upgrade pip
RUN /opt/isaacsim-env/bin/pip install isaacsim[all,extscache]==5.0.0.0 \
--extra-index-url https://pypi.nvidia.com
The two NVIDIA_* env vars are required – without them, the NVIDIA Container Runtime
does not inject the driver libraries and Vulkan/CUDA init will fail inside the container.
This was confirmed in a previous forum thread on the same setup.
You also need to add the Vulkan ICD and GLVND config files that the official image ships
(the EGL vendor JSONs under /usr/share/glvnd/egl_vendor.d/ and the Vulkan ICD JSON
under /etc/vulkan/icd.d/). See the official Dockerfile for the exact content (answer 2
below).
Run the container with --runtime=nvidia or --gpus all:
docker run --runtime=nvidia --gpus all -e NVIDIA_VISIBLE_DEVICES=all your-image ./your-script.py
Docs reference: Isaac Sim 5.0.0 – Python Environment Installation
Is the official Isaac Sim Dockerfile open-source?
Yes, as of Isaac Sim 6.0.0. It is published in the public GitHub repository:
tools/docker/Dockerfile
Note that this Dockerfile packages a pre-built Isaac Sim install (it does
COPY . /isaac-sim/), rather than doing a pip install at build time. It is still
useful as a reference for the exact system libraries, ENV vars, and Vulkan/EGL
configuration needed.
The accompanying Docker README
covers building, running, streaming, and Docker Compose deployment.
For Isaac Sim 5.0.0 specifically, the Dockerfile was not yet in the public repo at
release time; the container docs were the main reference:
Container Installation – Isaac Sim 5.0.0
What base image does the official Isaac Sim Docker image use?
The current official base (Isaac Sim 6.0.1):
FROM nvcr.io/nvidia/base/ubuntu:noble-20251013
noble is Ubuntu 24.04 (GLIBC 2.39). For a 5.0.0 pip install, Ubuntu 22.04 is
the better target since the 5.0.0 wheels require Python 3.11 and GLIBC 2.35, and
Ubuntu 22.04 satisfies both without needing a PPA. Ubuntu 24.04 is also fine (GLIBC 2.39
is compatible) but Python 3.11 is not the default there.
Closing this thread since the questions are answered. If you run into Vulkan or CUDA
init errors when building your image, the related thread
has additional troubleshooting. Feel free to open a new topic if you hit something
specific.