What happens when you press Ctrl + Alt + F1? That should take you to another terminal where you have to login with your username and password like this:
Nope. It’s not anything.
what about Ctrl+Alt+F2
Yeah that works but it just blocks the entire screen. Thats not a problem because I can use phone to talk with you. What I need to do next?
First, I would login and do nvidia-smi to see what that is giving you at the moment.
Then I would probably try rebooting and then after logging into the Ctrl+Alt+F2 terminal rerun these commands:
service gdm3 stop
sudo ./NVIDIA-Linux-x86_64-418.56.run
Make sure you are in the correct directory to run the .run file. Also maybe take pictures along the way so I can see if there is anything I am mis-communicating.
This is again the same error I got previously.
What does this show: ps a |grep X
try service gdm stop
It looks like something is still open so service gdm3 stop doesn’t seem to have worked.
Then run ps a |grep X again and see if that tty1 process is still there.
I believe the error you’re hitting there is due to the default compilation of the MKL-DNN support, which only works on x86-64, not ARM. This is noted in the build instructions here:
UPDATE 2019-02-07: MKL-DNN contraction kernels are now built by default.
MKL-DNN is compatible with x86_64 only so this needs to be disabled in the build preset.
tensorflow_bazel_options=--define=tensorflow_mkldnn_contraction_kernel=0has been added to the tensorflow_build_preset.patch.
I followed the patching instructions from that page and was able to successfully build an ARM64 toolchain on a Jetson Xavier using the Swift for TensorFlow snapshot as of 5/1/2019. I’ve uploaded this build to this location. It should let us work with the API as of the current 0.3 release, and my projects build and run using it on the Jetson Xavier and Nano (should be the same for the TX1 and TX2, but I don’t have the latest Jetpack installs on those).
This is still a CPU-only build, because I’m working on getting CUDA support in TensorFlow functional on the Jetson devices, but I can confirm that it runs on my Jetson Nano and also works correctly with Swift for TensorFlow Jupyter notebooks on that device. Others have described how to get CUDA-enabled TensorFlow builds working on the TX2, so I believe the same can be applied here to get this fully accelerated. It’s at least something that can be used for testing, even if it’s a little slow.
I tried another thing and now it seems to be lightdm
This feels like whack-a-mole. I am curious if it will now let you do service lightdm stop
Yeah now it’s just black screen with white cursor top left
can you ls or anything?
No can’t do anything. Should I press restart button on my computer?
Done! Back to square one.
ok, what does nvidia-smi give you at this point?
The same