Back to Skills

holoscan-install-debian

Install Holoscan SDK natively on Ubuntu via apt. Use for C++ installs on Ubuntu; pair with /holoscan-install-wheel for Python.

2,779stars322forksUpdated 8/4/2026

Security Assessment

Safe(92/100)
Security Score92/100

About holoscan-install-debian

holoscan-install-debian is an official NVIDIA install skill for installing the Holoscan SDK C++ runtime and headers natively on Ubuntu via apt. It solves the problem of choosing the correct holoscan-cuda-* package variant for the host's CUDA driver and configuring NVIDIA's apt repository, then verifying the install using the SDK's bundled C++ examples. Because apt does not provide Python bindings, the skill directs users needing Python to pair it with the /holoscan-install-wheel skill.

The workflow consults the official NVIDIA install docs first (authoritative on any disagreement), detects the Ubuntu version and CUDA driver via lsb_release and nvidia-smi, and applies a variant rule table to pick holoscan-cuda-13, holoscan-cuda-12, or holoscan (IGX). It checks prerequisites to skip redundant steps (cuda-keyring already installed, package already present), installs the cuda-keyring .deb and the chosen holoscan package via sudo apt-get, and then verifies by running the hello_world, tensor_interop, and video_replayer examples headless — patching the video_replayer YAML to stop after 10 frames. It handles the common 'CUDA driver version is insufficient' error by swapping the package variant. A skill-card and BENCHMARK.md document NVSkills-Eval results (security, correctness, discoverability, effectiveness, efficiency).

It targets developers and engineers building GPU-accelerated streaming AI applications who need a native Ubuntu C++ install of the Holoscan SDK. The operations are standard package management (wget of NVIDIA's signed cuda-keyring, sudo apt-get install) plus read-only verification runs, with no destructive or data-exfiltrating behavior.

FAQ

Which CUDA package variant should I install?

Follow the variant table by nvidia-smi CUDA version: 13.x+ uses holoscan-cuda-13, 12.x uses holoscan-cuda-12 (or plain holoscan on IGX). Installing the wrong variant produces a 'CUDA driver version is insufficient' error, which is fixed by removing and reinstalling the correct variant.

Does apt install give me Python bindings?

No. The apt install provides only the C++ runtime and headers; pair it with /holoscan-install-wheel if you need Python bindings.

What platforms are supported?

Ubuntu x86_64 (22.04 / 24.04) or ARM64 (Jetson / IGX) with an NVIDIA GPU and working driver. Other distributions must use the container or wheel install.

How is the install verified?

By running the bundled hello_world, tensor_interop, and video_replayer C++ examples, with video_replayer patched to run headless for 10 frames so it works over SSH without a display, and setting ulimit -s 32768.

What permissions does it need?

sudo and network access to developer.download.nvidia.com and docs.nvidia.com, primarily to add NVIDIA's signed cuda-keyring and run apt-get install — standard package-management operations.

All Files

5 files
evals/evals.json4.4 KB
View
skill-card.md3.7 KB
View
BENCHMARK.md3.8 KB
View
SKILL.md6.1 KB
View
skill.oms.sig4.5 KB
View

Install holoscan-install-debian

Download and extract the skill files to your .claude/skills/ directory.

Quick Setup:

  1. Copy the skill folder to .claude/skills/
  2. Claude will automatically detect and use the skill

Repository

nvidia/skills