SDK Installation

Set up your developer kit:

^{* iGPU and dGPU can be used concurrently on a single developer kit in dGPU mode. See details here.}

This version of the Holoscan SDK was tested on the Grace-Hopper SuperChip (GH200) with Ubuntu 22.04. Follow setup instructions here.

Supported x86_64 distributions:

¹ ^{Not formally tested or supported, but expected to work if building bare metal with the adequate dependencies.}
² ^{Not formally tested or supported, but expected to work if supported by the NVIDIA container-toolkit.}
³ ^{Not formally tested or supported, but expected to work if the glibc version of the distribution is 2.35 or above.}

NVIDIA discrete GPU (dGPU) requirements:

• Ampere or above recommended for best performance

• Quadro/NVIDIA RTX necessary for GPUDirect RDMA support

• Tested with NVIDIA Quadro RTX 6000 and NVIDIA RTX A6000

• NVIDIA dGPU drivers: 535 or above

• dGPU (x86_64, IGX Orin dGPU, Clara AGX dGPU, GH200)

  Copy

  Copied!

              
  
              
  docker pull nvcr.io/nvidia/clara-holoscan/holoscan:v2.3.0-dgpu
          

• iGPU (Jetson, IGX Orin iGPU, Clara AGX iGPU)

  Copy

  Copied!

              
  
              
  docker pull nvcr.io/nvidia/clara-holoscan/holoscan:v2.3.0-igpu
          

See details and usage instructions on NGC.

Try the following to install the holoscan SDK:

            

            
sudo apt update
sudo apt install holoscan
        

Troubleshooting

If holoscan is not found with apt:

            

            
E: Unable to locate package holoscan
        

Try the following before repeating the installation steps above:

• IGX Orin: Ensure the compute stack is properly installed which should configure the L4T repository source. If you still cannot install the Holoscan SDK, use the arm64-sbsa from the CUDA repository.

• Jetson: Ensure JetPack is properly installed which should configure the L4T repository source. If you still cannot install the Holoscan SDK, use the aarch64-jetson from the CUDA repository.

• GH200: Use the arm64-sbsa from the CUDA repository.

• x86_64: Use the x86_64 from the CUDA repository.

If you get missing CUDA libraries at runtime like below:

            

            
ImportError: libcudart.so.12: cannot open shared object file: No such file or directory
        

This could happen if your system has multiple CUDA Toolkit component versions installed. Find the path of the missing CUDA library (libcudart.so.12 here) using find /usr/local/cuda* -name libcudart.so.12 and select that path in sudo update-alternatives --config cuda. If that library is not found, or other cuda toolkit libraries become missing afterwards, you could try a clean reinstall of the full CUDA Toolkit:

            

            
sudo apt update && sudo apt install -y cuda-toolkit-12-2
        

If you get missing CUDA headers at compile time like below:

            

            
the link interface contains: CUDA::nppidei but the target was not found. [...] fatal error: npp.h: No such file or directory
        

Generally the same issue as above due from mixing CUDA Toolkit component versions in your environment. Confirm the path of the missing CUDA header (npp.h here) with find /usr/local/cuda-* -name npp.h and follow the same instructions as above.

If you get missing TensorRT libraries at runtime like below:

            

            
Error: libnvinfer.so.8: cannot open shared object file: No such file or directory
...
Error: libnvonnxparser.so.8: cannot open shared object file: No such file or directory
        

This could happen if your system has a different major version installed than version 8. Try to reinstall TensorRT 8 with:

            

            
sudo apt update && sudo apt install -y libnvinfer-bin="8.6.*"
        

If you cannot import the holoscan Python module:

            

            
ModuleNotFoundError: No module named 'holoscan'
        

To leverage the python module included in the debian package (instead of installing the python wheel), include the path below to your python path:

            

            
export PYTHONPATH="/opt/nvidia/holoscan/python/lib"
        

            

            
pip install holoscan
        

See details and troubleshooting on PyPI.