I am trying to download pytorch on a jetson NX to use a tracker i found, problem is that my conda is not able to find the correct cudatoolkit, i’ve trying downloading with pip but when running
torch.cuda.is_available()
i get false
i’ve tried downloading a separate cudatoolkit from another conda channel, and succeeded but when trying to install pytorch without cudatoolkit in the conda command, i get false again when trying to use cuda.
i tried using a different cudatoolkit version (no conda channel seem to have 10.2) and still false.
any help would be appreciated.
after downloading i get this:
print(torch.version)
1.8.1
print(torch.version.cuda)
None
torch.backends.cudnn.enabled
True
i saw a different topic suggesting removing cpuonly, my conda doesnt have it.
I created a completely new environment and did a clean downloaded and still nothing.
as it is a jetson i dont have nvcc or nvidia-smi i can only give you my cuda version which is 10.2.300 which came with the original build of the jetpack.