Manually installing CUDA, TensorRT, OpenCV, VisionWorks, etc.

I’m trying to install all of the post-flashing packages but the way our network is set up Jetpack just can’t handle SSH-ing into the device. Is there a way to manually install all of these just like you can manually flash?

Changing the network settings, setting up a local router, etc, is not an option.

You might find this of use:

Can you manually succeed at these:

ssh ubuntu@whatever_IP_address_it_is
ssh nvidia@whatever_IP_address_it_is
sudo ssh ubuntu@whatever_IP_address_it_is
sudo ssh nvidia@whatever_IP_address_it_is

The reason I ask is to see if it just wants a confirmation that you trust the connection. If that is the case, then you might still be able to use JetPack.

No, vanilla SSH commands don’t cut it with our network config. I can ssh in fine using the proper flags but Jetpack can’t

Is your network using a proxy? If manual command line connect doesn’t work there would be a problem. Knowing what “proper flags” are might suggest a workaround.

FYI, once JetPack downloads the “.deb” packages, if you do not delete the downloaded content, then those packages could be copied over by other means (e.g., scp or SD card). The trick is that they must be installed in the proper order. The CUDA repo package is always the first part to install.

I figured it could be manually installed I was just hoping someone else may have already done so I don’t have to figure out proper install order. I have the logs of a vanilla Jetpack install though so I’ll work off that as a starting point.

I suspect that if you first install the cuda repo package you won’t have too many problems…this allows “apt-get” to work (be sure to run “sudo apt update” first).

Normally dpkg itself can’t resolve dependencies. On the other hand, you might be able to cheat a bit by naming more than one dpkg in a single command and as a result have dpkg at least partially figure it out (I haven’t tried on dpkg, but it works on rpm and many package tools…give it a shot).