Clara Holoscan SDK
Build high-performance streaming AI applications for medical devices with the NVIDIA Clara™ Holoscan SDK. Access healthcare-specific acceleration libraries, pretrained AI models, and reference applications for ultrasound, endoscopy, surgical robots, and more.
DownloadLearn more about the Clara Holoscan Platform

Features
Reference AI Pipelines
.png)
Access reference architectures for endoscopy and ultrasound applications with video-in and AI inference for video-based modalities.
Sensor Processing

Support for camera serial interface and front-end sensors for video capture, ultrasound research, data acquisition, and connection to legacy medical devices.
Latency Measurement

Use Clara Holoscan SDK’s data transfer latency tool to measure complete end-to-end latency for video processing applications.
Compatible Platforms
- NVIDIA RTX support - Docker and CUDA, TensorRT, dGPU enablement
- NVIDIA Rivermax 100GbE streaming - Rivermax transport protocol streams data over ethernet directly into GPU GDDR DRAM using GPU Direct Technology
- NVIDIA Clara Guardian support - Build edge AI applications to improve patient care and operational efficiency using everyday sensors like cameras & microphones
Find the latest collection of reference applications on the NGC Collection for Clara Holoscan. In addition, find Deep Learning framework enabled Docker containers such as PyTorch and TensorFlow on NGC.
Download
1.
Apply for developer access to the Clara Holoscan SDK Program.
2.
Download and install the SDK Manager. After logging in to SDK Manager, you'll be able to flash your device.
3.
Download the Clara Holoscan SDK from GitHub.
Scroll down to the Documentation section to access user guides, NGC Collections, and sample code.
System Requirements
The Clara Holoscan SDK is compatible with the Clara AGX Developer Kit and Clara Holoscan Developer Kit. For more information on the SDK v0.2 platform and release information, please see the release notes in the Documentation section.
v0.1 | v0.2 | |
---|---|---|
Operating System | Ubuntu 18.04 | Ubuntu 20.04 |
Jetpack | Jetpack 4.5.1, l4t 32.5.1 | Jetpack 5.0 HP, l4t 34.1.2 |
Developer Kit | Clara AGX Developer Kit | Clara AGX Developer Kit, Clara Holoscan Developer Kit |
Supported GPUs | RTX 6000 | RTX 6000, RTX A6000 |
Supported Drivers | 460.73.01 | 510.73.08 |
Latest Version for Clara Holoscan SDK
The Clara Holoscan SDK v0.2 provides GXF integration, sample applications for endoscopy and ultrasound, GPUDirect RDMA with AJA integration, CUDA / OpenGL interoperability, TensorRT inference, and image transform acceleration in NVIDIA Performance Primitive Library (NPP). Visit the NGC Collection to run our Reference Applications today. To see the source code of our SDK, you can find it among a set of GitHub resources. If you have any questions or would like to exchange any ideas, visit the Developer Forum to chat with us.
Read More Reference Applications Forum
Documentation
Below you’ll find documentation for both the Clara AGX Developer Kit and Clara Holoscan SDK. If this is your first time getting started with either product, we suggest starting with the NVIDIA Clara AGX User Guide.
- NVIDIA Clara AGX Dev Kit User Guide
- NVIDIA Clara Holoscan SDK Documentation V0.2 (pdf) (html)
- NVIDIA Clara Holoscan Release Notes V0.2
Development Tasks | Getting Started |
Flash and set up the Clara AGX Developer Kit | Follow the Clara AGX Dev Kit User Guide. |
Connect your sensor IO to Clara AGX Developer Kit | Check out our distributor page for Holoscan IO Partners and their IO formats. |
Check out the reference applications | Visit the Clara Holoscan NGC Collection to see the run container for our GXF reference applications. |
Check out the source code | Visit the Clara Holoscan GitHub to find the source code for Clara Holoscan Embedded SDK. |
Explore the reference architecture for endoscopy use cases | Access the Endoscopy White Paper. |
Bring your own model to the endoscopy reference pipeline | Visit the Clara Holoscan SDK User Guide to see the Chapter Clara Holoscan Sample Apps on how to customize the reference apps. |
Use TensorRT for your deep learning inference applications | Visit the TensorRT developer page and getting started page for documentation and other resources on TensorRT. |
Develop drivers for GPU Direct/RDMA over PCIe as an ecosystem partner | Follow the minimal GPU Direct/RDMA driver sample code . |
Hardware
The NVIDIA Clara Developer kit helps facilitate accelerated processing with 3rd party sensor front ends, High-Speed I/O with the ConnectX SmartNIC, the flexibility and efficiency of the Jetson System-on-Module, and a powerful RTX-series Discrete GPU.
See our Clara Developer Kits Page for hardware options in your development journey, and the MGX product page for the production platform.
To find out more information about our ecosystem partners including IO partners and distributors, please visit our distributor page.
FAQ
It contains both. Clara Holoscan SDK is our software SDK that currently runs on the Clara AGX Developer Kit/Clara Holoscan Developer Kit hardware.
The Clara AGX/Holoscan Developer Kits are ARM based.
Scroll down to the Clara Holoscan Ecosystem Partners section in the distributor page .
The Clara AGX Developer Kit is the predecessor of the updated Clara Holoscan Developer Kit. Read more about our devkits here.
Clara AGX Developer Kit | Clara Holoscan Developer Kit | |
Availability | Now | Q4 2022 |
SoC | Xavier | Orin |
Discrete GPU (dGPU) | RTX 6000 | RTX A6000 |
SmartNIC | ConnectX-6 | ConnectX-7 |
Memory Bandwidth | 100 GbE / 10 GbE / 1 GbE | 2x 100 GbE / 2x 1GbE |
SSD | 250 GB | 500 GB |
Operating System | Ubuntu 18.04, Ubuntu 20.04 | Ubuntu 20.04 |
Jetpack Support | Jetpack 4.5, Jetpack 5.0 | Jetpack 5.0 |
Developer Tools | In dGPU mode: CUDA 11.1 | In dGPU mode: CUDA 11.6 |
Developer Blogs

June 17, 2022
Boost AI Medical Device Streaming Workflows with the Clara Holoscan SDK

March 22, 2022
New Sensor Partners Expand Surgical, Ultrasound, and Data Acquisition Capabilities in the NVIDIA Clara Holoscan Platform

February 2, 2022
Autonomous Robot Improves Surgical Precision Using AI

November 29, 2021