# System Requirements This page provides detailed hardware and software requirements to help set up and run the application efficiently. ## Hardware Requirements | **Component** | **Minimum** | **Recommended** | |---------------------|---------------------------------|--------------------------------------| | **Processor** | 11th Gen Intel® Core™ Processor | Intel® Core™ Ultra 7 Processor 155H | | **Memory** | 8 GB | 8 GB | | **Disk Space** | 256 GB SSD | 256 GB SSD | | **GPU/Accelerator** | Intel® UHD Graphics | Intel® Arc™ Graphics | ## Software Requirements - OS: Ubuntu 24.04.1 LTS (native installation, or as a WSL 2 distribution on Windows). - Docker Engine version 20.10 or higher. Docker Desktop is not supported on Linux, because its virtual machine cannot access the host `/dev/dri` render nodes required for GPU acceleration. - Docker Compose version 2.34.0 or newer. - For GPU and/or NPU usage, appropriate drivers must be installed. The recommended method is to use the DL Streamer installation script, which detects available devices and installs the required drivers. Follow the **Prerequisites** section in [DL Streamer Install Guide - Ubuntu](https://docs.openedgeplatform.intel.com/2026.3/edge-ai-libraries/dlstreamer/install/install_guide_ubuntu.html#prerequisites). ## Network Requirements **An outbound internet connection is required.** ViPPET is not supported in air-gapped or fully offline environments: container images, sample videos, models, and Python packages are all fetched on demand and are not bundled with the tool. | **When** | **What is downloaded** | **From** | | ---------------------- | --------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | | Installation | Docker Engine, GPU/NPU drivers, `git`, `make`, `curl` | Ubuntu and Docker apt repositories | | Installation | ViPPET repository sources | `github.com` | | First start | Pre-built container images | `docker.io` (Docker Hub) | | First start | `model-download` plugin virtual environments (Python packages) | `pypi.org`, `files.pythonhosted.org` | | First start | Default sample recordings listed in `shared/videos/default_recordings.yaml` | `storage.openvinotoolkit.org`, `github.com`, `pexels.com` | | Model installation | Model weights and metadata for the selected hub | `huggingface.co`, `ultralytics.com`, `storage.openvinotoolkit.org`, Intel® Geti™ | | Build from source | Base images, apt packages, Python and npm dependencies | `docker.io`, distribution and language package registries | | UI and API docs in use | Web fonts and the Swagger UI bundle | `fonts.googleapis.com`, `cdn.jsdelivr.net` | > [!NOTE] > > - Downloaded artifacts are cached under `shared/`, so subsequent starts need far less bandwidth. A connection > is still needed whenever a new model or sample video is installed, or after `make clean`. > - The first start can take several minutes and download several GB, depending on the selected models. > - Behind a corporate proxy, export `http_proxy`, `https_proxy`, and `no_proxy` in the shell before starting > the application, and configure the [Docker daemon proxy](https://docs.docker.com/engine/daemon/proxy/) so that > image pulls succeed. `compose.yml` already appends the internal service names to `no_proxy`. > - Access to the Hugging Face Hub additionally requires a token for gated or private repositories. See > [Pre-Installation Steps](./pre-installation-steps.md). > - Inbound access is not required. The UI is served on port `80` and is reachable at `http://localhost` or > `http://` on the local network. ## Windows Subsystem for Linux (WSL) Ubuntu 24.04 running under WSL 2 on Windows is supported. The installation steps are identical to a native Ubuntu installation - run all commands ([Use Pre-Built Docker Images](./docker-compose.md) or [Build from Source](./build-from-source.md)) inside the Ubuntu 24.04 WSL distribution. `setup_env.sh` checks NPU devices first, then `/dev/dxg`, then native `/dev/dri/render*` nodes, and finally falls back to CPU. On WSL, `/dev/dxg` selects the `gpu-wsl` Compose profile even when `/dev/dri/render*` also exists. The profile passes `/dev/dxg` and mounts `/usr/lib/wsl` read-only for the ViPPET backend, DLStreamer Pipeline Server, Metrics Manager, and ASR service through `compose.gpu-wsl.yml`. Metrics Manager's native GPU device mapping is replaced rather than merged, so `/dev/dri` is not required by the standard WSL configuration. Experimental Time Series Analytics uses `compose.experimental.yml` without a WSL-specific override. It still requires `/dev/dri` and a valid `RENDER_GROUP_ID`; `/dev/dxg` alone is not sufficient for that service. The Windows Intel graphics driver must support GPU compute under WSL, and `/dev/dxg` and `/usr/lib/wsl` must be available inside the WSL distribution. GPU telemetry from native Linux collectors such as qmassa may be unavailable under WSL; a missing GPU utilization chart does not prove that inference is running on CPU. Run `make run` to regenerate `.env` and apply the detected profile. When invoking Compose directly, use `COMPOSE_PROFILES=gpu-wsl` and `compose.gpu-wsl.yml`. Do not edit `.env` manually. The runtime device is still `GPU`, and pipeline labels remain **GPU (WSL)**. ### Supported pipeline variants under WSL | **Variant** | **Supported under WSL** | |---------------|-------------------------| | CPU | Yes | | GPU (WSL) | Yes | | GPU (native) | No | | NPU | No | Pipelines expose a dedicated **GPU (WSL)** variant that is shown only when the application runs under WSL. Native GPU and NPU variants are hidden in that environment.