The current version of NVIDIA Isaac Sim is 6.2.0. To run it you need at least an RTX 4080-class GPU with 16 GB of memory, 32 GB of RAM, 50 GB of free SSD space and Ubuntu 22.04, Ubuntu 24.04 or Windows 11 (Windows 10 isn’t supported). You can install it as a standalone download, with pip on Python 3.12, as a Docker container or in the cloud. ROS 2 is a separate setup from Isaac Sim itself: NVIDIA recommends ROS 2 Jazzy on Ubuntu 24.04 and documents a native Jazzy setup on Windows 11. It’s free for internal research and development; you need an NVIDIA AI Enterprise license only to redistribute it or deliver it to clients.
System requirements (Isaac Sim 6.2.0)
| Minimum | Good | Ideal | |
|---|---|---|---|
| GPU | RTX 4080 (16 GB) | RTX 5080 (16 GB) | RTX PRO 6000 Blackwell (48 GB) |
| CPU | Core i7 7th gen / Ryzen 5, 4 cores | Core i7 9th gen / Ryzen 7, 8 cores | Core i9 X / Ryzen 9 or Threadripper, 16 cores |
| RAM | 32 GB | 64 GB | 64 GB |
| Storage | 50 GB SSD | 500 GB SSD | 1 TB NVMe |
From NVIDIA’s requirements page, last updated October 14, 2026. The tested drivers are 595.58.03 on Linux and 595.97 on Windows. On Arm, Isaac Sim runs on DGX Spark with DGX OS 7. Ubuntu 22.04 still works in 6.2 but is marked deprecated, so start new machines on 24.04.
Four ways to install
- Standalone download (Linux and Windows 11): unzip the 6.2.0 package, run
post_install.sh, thenisaac-sim.sh(or the .bat files on Windows). Best if you want the full desktop app. - pip (Linux and Windows): on Python 3.12, with GLIBC 2.35 or newer on Linux (on Windows the guide notes you may need to enable long path support), install PyTorch 2.11 first, then
pip install isaacsim[all,extscache]==6.2.0.0 --extra-index-url https://pypi.nvidia.com, per the Python install guide. Best if Isaac Sim is one part of a Python training project. - Docker (Linux host; NVIDIA’s steps use Ubuntu):
docker pull nvcr.io/nvidia/isaac-sim:6.2.0, per the container guide. Best for headless servers and teams. Its web viewer is supported on Ubuntu hosts and DGX Spark only, not on Windows hosts or WSL. - Cloud: NVIDIA’s quick-install page links guides for AWS, Azure, GCP and other providers, useful when you don’t have a 16 GB GPU locally.
On Windows 11 you install Isaac Sim itself with the standalone download or pip. Pixi is a separate step: NVIDIA uses it to set up the external ROS 2 environment on Windows, covered in the next section.
ROS 2: which distribution goes with which OS
Isaac Sim ships its own ROS libraries for the default interfaces; the ROS 2 install you set up is for your external nodes. NVIDIA’s ROS 2 setup page covers these combinations:
| OS | ROS 2 | How the external ROS is set up | Status on NVIDIA’s page |
|---|---|---|---|
| Ubuntu 24.04 | Jazzy | apt packages (or source build) | Recommended |
| Ubuntu 24.04 | Humble | — | Not supported (stated) |
| Ubuntu 22.04 | Humble | apt packages | Documented default (Ubuntu 22.04 itself is deprecated in Isaac Sim) |
| Ubuntu 22.04 | Jazzy | Docker container or source build | Documented (no Jazzy apt packages for 22.04) |
| Windows 11 | Jazzy | Pixi, native | Documented procedure |
| Windows 11 | Humble | Pixi, native | Not verified in NVIDIA’s docs (not stated as unsupported) |
| Windows 11 | Humble or Jazzy | WSL2 | Deprecated; custom interfaces not supported |
Isaac Sim itself runs Python 3.12, while your external ROS nodes keep their own Python version in a separate terminal. Custom message packages have to be built for the simulator’s Python 3.12 runtime (NVIDIA provides build_ros.sh for this). If you’re new to ROS 2, our first ROS 2 steps without a robot come first.
What changed in Isaac Sim 6
- 6.0 (general availability announced June 8, 2026): support for the Newton physics engine in the URDF and MJCF importers, native RTX lidar full-scan accumulation, an acoustic sensor, a web streaming client in the container, native Windows install, and an MCP server so AI coding tools can drive the simulator.
- 6.1: Newton 1.5 with new solvers and hydroelastic contact, a ros2_control extension for Humble and Jazzy, Clearpath robot models and SO-101 system-identification tools, and 1,533 SimReady assets.
- 6.2, per the release notes: Newton 1.6.1, MuJoCo/MJWarp 3.12, Warp 1.17, refreshed Windows ROS 2 packages and Isaac ROS 5.0 compatibility. The old Cortex framework was removed, so move any Cortex code before upgrading.
Isaac Sim and Isaac Lab
Isaac Sim is the simulator; Isaac Lab is the training framework that runs on top of it. NVIDIA’s compatibility table pairs Isaac Sim 6.2.0 with Isaac Lab 3.0.0 and 6.1.0 with Isaac Lab 3.0 early access. Install the pairs NVIDIA lists; other combinations are outside its compatibility table, and version mismatches can show up as import errors. Our Isaac Lab getting-started guide covers the training side, and our checks on camera observations and demonstration replay cover two common pitfalls.
The license, in plain terms
NVIDIA’s license FAQ says:
- The source on GitHub is Apache 2.0; the Kit SDK and assets have a separate NVIDIA license.
- Isaac Sim is free for internal research and development, with no seat limit.
- You can sell what you make with it (trained models, code, USD assets) without a license.
- You need an NVIDIA AI Enterprise license to redistribute Isaac Sim, offer it as a service to others, or install it as part of a turnkey system on a client’s hardware.
Isaac Sim or another simulator?
Isaac Sim is the heaviest option, and the right one when you need photorealistic sensors, synthetic data or NVIDIA’s robot learning tools. For lighter ROS work, Gazebo runs on modest hardware; see our Gazebo version guide and the Isaac Sim vs MuJoCo vs Gazebo comparison. For how Isaac Sim fits with Cosmos, GR00T and Jetson Thor, read our NVIDIA physical AI guide.
NVIDIA Isaac Sim documentation (requirements, install and ROS 2 pages) checked October 24, 2026.