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/drirender 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.
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, |
Ubuntu and Docker apt repositories |
Installation |
ViPPET repository sources |
|
First start |
Pre-built container images |
|
First start |
|
|
First start |
Default sample recordings listed in |
|
Model installation |
Model weights and metadata for the selected hub |
|
Build from source |
Base images, apt packages, Python and npm dependencies |
|
UI and API docs in use |
Web fonts and the Swagger UI bundle |
|
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 aftermake 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, andno_proxyin the shell before starting the application, and configure the Docker daemon proxy so that image pulls succeed.compose.ymlalready appends the internal service names tono_proxy.Access to the Hugging Face Hub additionally requires a token for gated or private repositories. See Pre-Installation Steps.
Inbound access is not required. The UI is served on port
80and is reachable athttp://localhostorhttp://<HOST-IP>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 or Build from Source) 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.