Pytorch Tensorflow Support in Fedora 40 with NVIDIA GPU

This project took some time to get working, what it involved was getting the GPU acceleration for the two major frameworks

go to this link: to install drivers

then install cockpit

sudo dnf install cockpit cockpit-podman podmancockpit systemctl enable --now cockpit.socket

Open a web browser and go to http://localhost:9090

and go to command prompt in cockpit, make sure NVIDIA container toolkit is installed

podman run --device nvidia.com/gpu=all -p 8888:8889 -it docker.io/pytorch/pytorch:latest

For Pytorch

 podman run --device nvidia.com/gpu=all -p 8888:8889 -it nvcr.io/nvidia/tensorflow:24.01-tf2-py3


#for tensorflow for GPU inside the pod, install jupyter
pip install jupyter
#to get jupyter running, Make sure to make a firewall port 8888 and 8889 in the cockpit
jupyter notebook --ip=0.0.0.0 --port=8889 --no-browser --allow-root
#connect to jupyter through link 8888 if that doesn't work (it's finecky), and put in your token and log in. To test tensorflow #TORCH:

import torch 
def check_gpu():
    if torch.cuda.is_available():
        print("CUDA is available. Number of GPUs available:", torch.cuda.device_count())
        print("Details of each GPU:")
        for i in range(torch.cuda.device_count()):
            print(f"GPU {i}: {torch.cuda.get_device_name(i)}")
            print(f" Memory Allocated: {torch.cuda.memory_allocated(i)} bytes")
            print(f" Memory Cached: {torch.cuda.memory_reserved(i)} bytes")
       else:
           print("CUDA is not available. No GPU detected.") 
check_gpu()
#TENSORFLOW:

import tensorflow as tf 
def check_tf_gpu():
    if tf.config.list_physical_devices('GPU'):
       print("TensorFlow is using the GPU(s).")
       gpus = tf.config.list_physical_devices('GPU')
    for gpu in gpus:
        print(f"GPU: {gpu.name}")
   else:
      print("TensorFlow is not using the GPU.") 
check_tf_gpu()