The Jetson Nano can run this code with a small change to the cmake options

[SMSmith]

(DON'T MERGE THIS - see below)
By removing compute_86, this code compiles and runs with the following library versions:

• Jetpack-4.7.2
• CUDA-10.2
• CUDNN-8.2.1
• TensorRT-8.2.1
• OpenCV-4.7.0 (you could probably use an earlier version)
• CMake-3.26.3 (you could probably use an earlier version)

You need gcc-8/g++-8 or better. Onnxsim requires onnxoptimizer, which struggles to compile on the nano, recommend making sure everything else is closed before installing. Use a lightdm or similar window manager if you need one to cut down on background memory usage and aren't going headless.

Is there a better way to make the CMakelists only use compute_86 if the hardware isn't the nano? Or add a nano option to make? Actually deleting the compute_86 is probably detrimental to performance on other platforms. (AKA DON'T MERGE THIS)

[triple-Mu]

Thanks for sharing such job.
Sure we also want to support sm80 89 and so on.