SDK 4.2.0 CUDA 11.x installer doesn't work

Number of problems.

First of all it says:

Checking CUDA version...
OK: Found CUDA 11.5
Driver recently setup, reboot your computer to use CUDA

But I’ve already rebooted my machine since installing CUDA. So it’s detecting something incorrectly.

More importantly, it says:

Install samples (recommended) [Y/n] ?n
cp: target '/usr/local/cuda/lib64/' is not a directory
cp: target '/usr/local/cuda/include/' is not a directory
chmod: cannot access '/usr/local/cuda/include/': No such file or directory
chmod: cannot access '/usr/local/cuda/lib64/*': No such file or directory
ls: cannot access '/usr/local/cuda/lib64/libnvinfer_builder_resource.so.8.6.1*': No such file or directory
basename: missing operand
Try 'basename --help' for more information.

But I’m pretty sure CUDA is installed in /usr/lib/cuda and /usr/include/cuda, not /usr/local/cuda/lib or /usr/local/cuda/include. So it thinks CUDA is in the wrong place.

Then it says:

Do you want to run the ZED Diagnostic to download all AI models [Y/n] ?y

Cannot find CUDNN

But I have CUDNN installed.

Then it fails to download models and the SDK does not function correctly.

It looks to me like CUDA isn’t installed properly. Please check if the CUDA version is displayed using the command “nvcc --version or nvcc -V”. If it’s not displayed properly, please also check with “nvidia-smi”. I don’t know if it will help, but it would be good to check!

And if you are using an embedded board, it is convenient to initially set up CUDA and cudnn through the SDK Manager.