Clara AGX Developer Kit User Guide
{iGPU,dGPU} install iGPU or dGPU driver stack
optional arguments:
-h, --help show this help message and exit
-f, --force force reinstallation of the specified driver stack
-d, --dry do a dry run, showing the commands that would be
executed but not actually executing them
-i, --interactive run commands interactively (asks before running each
command)
-v, --verbose enable verbose output (used with --dry to describe
the commands that would be run)
-l LOG, --log LOG writes a log of the install to the specified file
-r [L4T_REPO], --l4t-repo [L4T_REPO]
specify the L4T apt repo (i.e. when using an apt
mirror; default is repo.download.nvidia.com/jetson)
4.
The dGPU driver install may be verified using the
query
command:
$ nvgpuswitch.py query
dGPU (cuda-drivers, 510.73.08-1)
Quadro RTX 6000, 24576 MiB
5.
After the dGPU drivers have been installed, rebooting the system will complete the switch
to the dGPU. At this point the Ubuntu desktop will be output via DisplayPort on the dGPU,
and so the display cable must be switched from the onboard HDMI (port 10) to DisplayPort
(port 4) on the dGPU.
Note: If the output connection isn’t switched before the devkit finishes rebooting, the
terminal screen will hang during booting.
6.
Modify PATH and LD_LIBRARY_PATH . CUDA installs its runtime binaries such as nvcc into
its own versioned path that is not included by the default $PATH environment variable.
Because of this, attempts to run commands like nvcc will fail on dGPU unless the CUDA
11.6 path is added to the $PATH variable. To add the CUDA 11.6 path for the current user,
add the following lines to $HOME/.bashrc after the switch to dGPU:
export PATH=/usr/local/cuda-11.6/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.6/lib64:$LD_LIBRARY_PATH
At this time, the Clara Holoscan SDK is tested and supported only in dGPU mode. Switching
back to iGPU mode after switching to dGPU mode is not recommended.
Note: The GPU settings will persist through reboots until it is changed again with
nvgpuswitch.py
.
Reinstalling Optional SDK Packages
This section only applies if you have selected “Additional SDKs” in Step 01 of the SDK Manager
installation process.
When switching between GPUs, CUDA is first uninstalled and then reinstalled by the script in
order to provide the correct versions used by iGPU or dGPU (CUDA 11.4 and 11.6,
respectively). Since some optionally installed packages via SDK Manager such as DeepStream