Run yolov8 inference on Jetson nano unable to recognize gpu

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Is there any known incompatibility issue regarding yolov8 (8. 0.116) requirements and jetson nano set up? When running yolo command line with device=0 , It doesn’t seem to recognize cuda device . I have cuda 10.2 and pytorch 1.9.0 with torchvision 0.10.0 . Python3.7.5, Opencv 4.1.1 and Ubuntu 18.04. Nvcc --version command returns the cuda version and i have added the PATH for /usr/local/cuda/bin including the library, in .bashrc file. However torch.cuda.is_available() returns ‘False’…
Please note that i was able to run yolov5 without problems last year with python3.6 i think or lower.

3

Hi,

How do you install the PyTorch package?
Do you use our prebuilt package?

https://docs.nvidia.com/deeplearning/frameworks/install-pytorch-jetson-platform/index.html

Thanks.

4

Thanks for your prompt response. No I did not install it exactly this way. Maybe I will uninstall and install following this link and putting python 3.7. Instead of python3 and torch==1.9.0, correct?
FYI torch was already installed from my previous project with yolov5 which worked but now I had to change python, upgrade to 3.7.5 in order for the yolov8 to work. Pip install Ultralytics also installs torch 1.13, so afterwards I pip install torch==1.9.0… But still cuda wasn’t recognized. Then I tried to uninstall torch and install again using pip install torch==1.9.0… Note that pip show torch and import torch work correctly showing torch 1.9.0 and the corresponding file.

Thank you,
Elizabeth Bellou

Στις Δευ, 12 Ιουν 2023, 11:00 ο χρήστης AastaLLL via NVIDIA Developer Forums <notifications@nvidia.discoursemail.com> έγραψε:

5

Hi,

The package from pip installer is the CPU version package.
Please install our prebuilt which is located at Index of /compute/redist/jp.

Do you use Nano or the new Orin Nano?
For the 1st Nano which uses JetPack 4, we only provide the prebuilt for python 3.6.

Thanks.

6

I use Jetson nano 4gb with jetpack 4.6

Thank you,
Elizabeth Bellou

Στις Τρί, 13 Ιουν 2023, 08:20 ο χρήστης AastaLLL via NVIDIA Developer Forums <notifications@nvidia.discoursemail.com> έγραψε:

7

Hi,

The latest software for Nano is JetPack 4.6.3.
You can try the installation command below:

$ export TORCH_INSTALL=https://developer.download.nvidia.com/compute/redist/jp/v461/pytorch/torch-1.11.0a0+17540c5+nv22.01-cp36-cp36m-linux_aarch64.whl
$ python3 -m pip install --no-cache $TORCH_INSTALL

Thanks.

8

Thank you for your help, by installing torch this way, I managed to get torch.cuda.is_available() True , however only with python 3.6.9 … Yolov8 run with python >=3.7. I tried installing pytorch with python 3.7.5 there was an error that it is not supported for this platform. You had mentioned this incompatibility for jetpack 4 in your previous answer. Maybe I should reach yolov8 develooers forum.

Thank you,
Elizabeth Bellou

Στις Τετ, 14 Ιουν 2023, 06:02 ο χρήστης AastaLLL via NVIDIA Developer Forums <notifications@nvidia.discoursemail.com> έγραψε:

10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.