SDK Installation
The section below refers to the installation of the Holoscan SDK referred to as the development stack, designed for NVIDIA Developer Kits based on HoloPack or JetPack, and for x86_64 Linux compute platforms, ideal for development and testing of the SDK.
For Holoscan Developer Kits such as the IGX Orin Developer Kit, an alternative option is the deployment stack, based on OpenEmbedded (Yocto build system) instead of Ubuntu. This is recommended to limit your stack to the software components strictly required to run your Holoscan application. The runtime Board Support Package (BSP) can be optimized with respect to memory usage, speed, security and power requirements.
Set up your developer kit:
Developer Kit |
User Guide |
L4T |
GPU Mode |
|---|---|---|---|
| NVIDIA IGX Orin | Guide | HoloPack 2.0 | iGPU or* dGPU |
| NVIDIA Clara AGX | Guide | HoloPack 1.2 | dGPU* |
| NVIDIA Jetson AGX Orin | Guide | JetPack 5.1.1 | iGPU |
* iGPU and dGPU can be used concurrently on a single developer kit in dGPU mode with the L4T Compute Assist container
For ConnectX support on the IGX Orin and Clara AGX developer kits, install the MOFED drivers (>=5.8). They can also be installed by selecting the Rivermax option in SDK Manager, available if you are part of the Rivermax SDK program.
If the developer kit is in dGPU mode, reconfigure its GPUDirect driver after installing MOFED to enable RDMA (link).
You’ll need the following to use the Holoscan SDK on x86_64:
OS: Ubuntu 20.04
NVIDIA discrete GPU (dGPU)
Ampere or above recommended for best performance
Quadro/NVIDIA RTX necessary for GPUDirect RDMA support
Tested with NVIDIA RTX 6000 and NVIDIA RTX A6000
NVIDIA dGPU drivers: 510.73.08 or above
For ConnectX support (optional):
Additional software dependencies might be needed based on how you choose to install the SDK (see section below).
Refer to the Additional Setup and Third-Party Hardware Setup sections for additional prerequisites.
We provide multiple ways to install and run the Holoscan SDK:
| Runtime libraries | Included | Included | Included |
| Python module | 3.8 | 3.8 | 3.8 to 3.11 |
| C++ headers and CMake config |
Included | Included | N/A |
| Examples (+ source) | Included | Included | retrieve from GitHub |
| Sample datasets | Included | retrieve from NGC |
retrieve from NGC |
| CUDA 11 runtime 1 | Included | automatically 2 installed |
require manual installation |
| NPP support 3 | Included | automatically 2 installed |
require manual installation |
| TensorRT support 4 | Included | automatically 2 installed |
require manual installation |
| Vulkan support 5 | Included | automatically 2 installed |
require manual installation |
| V4L2 support 6 | Included | automatically 2 installed |
require manual installation |
| Torch support 7 | Included | require manual 8 installation |
require manual 8 installation |
| Rivermax support 9 | add on top 10 of the image |
require manual 11 installation |
require manual 11 installation |
| CLI support | needs docker dind with buildx plugin on top of the image |
needs docker w/ buildx plugin |
needs docker w/ buildx plugin |
Refer to the documentation in each of these for specific install and run instructions.
Not sure what to choose?
The Holoscan container image on NGC it the safest way to ensure all the dependencies are present with the expected versions (including torch 8). It is the simplest way to run the embedded examples, while still allowing you to create your own C++ and Python Holoscan application on top of it. These benefits come with the standard inconvenience that exist when using Docker, such as more complex run instructions for proper configuration. Also, supporting Rivermax or the CLI require more work than the other solutions at this time.
If you are confident in your ability to manage dependencies on your own in your host environment, the Holoscan Debian package should provide all the capabilities needed to use the Holoscan SDK.
If you are not interested in the C++ API but just need to work in Python, or want to use a newer version than Python 3.8 (up to 3.11), you can use the Holoscan python wheels on PyPI (just
pip install holoscan). While they are the easiest solution to get started, you might need additional work to setup your environment with adequate dependencies depending on your needs.
Need more control over the SDK?
The Holoscan SDK source repository is open-source and provides reference implementations as well as infrastructure for building the SDK yourself.
We only recommend building the SDK from source if you need to build it with debug symbols or other options not used as part of the published packages. If you want to write your own operator or application, you can use the SDK as a dependency (and contribute to HoloHub). If you need to make other modifications to the SDK, file a feature or bug request.
CUDA 11.4+ (< 12.0) is required. Already installed on NVIDIA developer kits with HoloPack and JetPack.
Debian installation on x86_64 requires the latest cuda-keyring package to automatically install all dependencies
NPP 11.4+ needed for the FormatConverter operator. Already installed on NVIDIA developer kits with HoloPack and JetPack.
TensorRT 8.2.3+ and cuDNN needed for the Inference operator. Already installed on NVIDIA developer kits with HoloPack and JetPack.
Vulkan 1.2.131+ loader needed for the HoloViz operator (+ libegl1 for headless rendering). Already installed on NVIDIA developer kits with HoloPack and JetPack.
V4L2 1.18+ needed for the V4L2 operator. Already installed on NVIDIA developer kits with HoloPack and JetPack.
Torch support requires LibTorch 1.12+, TorchVision 0.14.1+, OpenBLAS 0.3.8+ (all systems) and OpenMPI (aarch64 only)
To get LibTorch and TorchVision on aarch64 (NVIDIA devkits), build them from source, download our pre-built installation, or copy them from the holoscan arm64 container (in /opt).
Tested with Rivermax SDK 1.20
Supporting the Rivermax SDK in a container also requires adding the Mellanox OFED user space in that container
Rivermax SDK and OFED drivers are installed on NVIDIA developer kits with [SDKM][sdkm] though the [Rivermax SDK program][rivermax-program]