CUDA is installed and verified on Jetson TX/NX. Also Pytorch with CUDA is installed using pip. But while running Detectron2 the following error comes up - “AssertionError: Torch not compiled with CUDA enabled”. Also this function - torch.cuda.is_available() returns false when called.
Seems to be some Jetpack setup issue.
Which jetpack version is being used in this case? Is this issue with Jetpack 5.0?
Are you guys facing the same issue with Jetpack 4.6.1 as well?
I don’t think TX2 series is currently supported in 5.x version in case you are using Jetpack 5.x.
“This release supports Jetson AGX Orin Developer Kit as well as Jetson AGX Xavier series and Jetson Xavier NX series developer kits and commercial modules.”
We have tried on Jetpack version 4.6-b199. The use case that we are working on involves detectron2 which requires python version of 3.7 or above. Detectron2 doesn’t work with pytorch compiled with python version 3.6. We need support to get the pytorch working with python 3.9 which will detect GPU on the Jetson TX.