NVIDIA Jetson AGX Thor - Holoscan

Follow Us On

Share This Page

Facebook Twitter LinkedIn WhatsApp Email Copy Link

Close

NVIDIA®Jetson AGX Thor™
Introduction
SoM Overview Carrier Boards NVIDIA Developer Kit Auvidea X241 Connect Tech Gauntlet
JetPack 7.0
Getting Started Installing JetPack Flashing the Board with GUI Installer JetPack Components Compiling Kernel and DTB Flashing the Board from Command Line Flashing the Board from USB Performance Tuning Evaluating Performance Set Values Manually Tuning Power Maximizing Performance
Development in the Board
Board Access Methods Using Serial Console Using SSH Using Ubuntu Login Installing Packages
Camera Support
Camera Over Ethernet (CoE) MIPI CSI IMX477 + J20 Driver
Blackwell GPU
GStreamer Pipelines
Capture and Display H.264 H.265
Holoscan
PVA
DeepStream
Installation Running Examples Performance Benchmarking
RidgeRun Products
GstCUDA RTSP Server Video Stitching Video Stabilization Bird's‑Eye View Qt Overlay GStreamer Browser Sink Development Suite Spherical Video PTZ
RidgeRun Open Source Projects
GstPerf GStreamer Daemon GstInterpipe GstShark
RidgeRun Reference Designs
Panoramic Stitching and WebRTC Streaming
Reference Documentation
Contact Us

Overview

NVIDIA Holoscan is an AI sensor processing platform designed to enable low-latency streaming AI pipelines across a range of domains, including medical devices, edge high-performance computing, and industrial inspection. It integrates specialized hardware, optimized software libraries, and microservices to support applications spanning from embedded systems to edge and cloud environments.

Core Features:

1. Flexible installation: offers multiple installation strategies
   • Containers
   • Python wheels
   • Source builds
2. APIs for development:
   Holoscan provides both C++ and Python APIs, which are the recommended interfaces for building application pipelines. Developers can use these to implement custom operators, build workflows, and integrate AI models.
3. Built-in Operators: basic units of work within Holoscan applications. The SDK includes domain-agnostic operators for tasks such as:
   • Input/Output
   • Machine learning inference
   • Processing
   • Visualization
4. Examples and Tutorials:
   The SDK provides minimal examples with step-by-step explanations in the Holoscan by Example section. Developers can also explore example source code in the GitHub repository.
5. HoloHub Repository: open-source repository where users share:
   • Reusable operators
   • Sample applications
6. Packaging and Deployment: The Holoscan CLI simplifies packaging and deployment of applications into OCI-compliant containers, which are suitable for production use cases.
7. Performance Tools: Holoscan integrates tools to profile and validate application performance, including:
   • Data Flow Tracking: Profiling data flows between operators.
   • Performance benchmarking to ensure low latency and high throughput.

Core Technologies Powering Holoscan

1. Rivermax & GPUDirect RDMA: High-speed networking optimized for GPU workloads with zero-copy transfers to pinned GPU memory.
2. Graph Execution Framework (GXF): Enables efficient execution of task pipelines while minimizing memory copies.
3. TensorRT: Provides optimized deep learning inference on NVIDIA GPUs.
4. CUDA + Vulkan Interoperability: Efficient sharing of GPU resources for real-time rendering and compute.
5. NVIDIA NPP: CUDA-accelerated library for common image transformations.
6. UCX (Unified Communications X): High-performance framework for distributed applications with multiple transport layer supports.
7. MatX: C++17 GPU-accelerated numerical computing library with seamless integration into Holoscan.

Getting Started with Holoscan

1. Choose your platform: Supported on NVIDIA Developer Kits (aarch64) and x86_64 workstations.
2. Install the SDK: Flexible setup options with additional configurations for performance and external sensors.
3. Learn the framework:
4. Study the Core Concepts.
5. Follow tutorials in Holoscan by Example or consult condensed API documentation.
6. Explore reusable operators: Built-in SDK operators and HoloHub components can speed up development.
7. Write and run your application: Use provided utilities for packaging and distribution.
8. Master advanced details: Dive deeper into logging, data flow tracking, and GXF execution internals.

Installation guide

This section refers to the installation of the holoscan SDK referred to as the development stack, designed for NVIDIA Developer Kits (arm64) and for x86_64 Linux compute platforms. There are multiple ways to install and run the Holoscan SDK.

Container image on NGC it the safest way to ensure all the dependencies are present with the expected versions (including Torch and ONNX Runtime) and should work on most Linux distributions. Still allowing you to create your own C++ and Python Holoscan application. If you manage dependencies correctly on your own in your host environment, the Holoscan Debian Package should provide all the capabilities needed, assuming you are on Ubuntu 22.04 or 24.04
If you are not interested in C++ API but just need to work with Python you can use Holoscan Python wheels on PyPI. While they are the easiest solution to install the SDK, it might require the most work to setup your environment with extra dependencies based on your needs. Finally, they are only formally supported on Ubuntu 22.04 and Ubuntu 24.04

NGC Container

docker pull nvcr.io/nvidia/clara-holoscan/holoscan:v3.6.0-dgpu

Debian package

sudo apt update
sudo apt install holoscan

This will not install dependencies needed for the Torch nor ONNXRuntime inference backends. To do so, install transitive dependencies by adding the --install-suggests flag to apt install holoscan

Python Wheel

pip install holoscan