Different GPU detection between python command line and Jupyter Notebook Cell

[fiammante]

Describe the bug
Tensorflow GPU support existing in command line python is not available in Jupyter notebooks started from same conda environment.

To Reproduce
Follow the steps to install Tensorflow GPU support on Windows WSL (using UBUNTU 22.04.2 LTS distro)
https://www.tensorflow.org/install/pip#windows-wsl2

The verification command
python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"
returns
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

Install jupyter in the miniconda environment defined in the tensorflow install page.

Launch jupyter notebook.exe from the same command line environment
Same Python code from a notebook cell
import tensorflow as tf;
print(tf.config.list_physical_devices('GPU'))

return and empty list []

Expected behavior
Notebook cell should display same result as command line [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
Screenshots

Desktop (please complete the following information):

• OS: WSL2 under Windows 10 latest update using Ubuntu 22.04.2 LTS distro
• Browser Firefox
• Version Firefox 113.0.2 (64 bits)

Additional context
the jupyter command help under Ubuntu states to use jupyter notebook.exe (not jupyter notebook).
same behavior after installing nb_conda.

[RRosio]

Hi @fiammante thank you for submitting this issue! Can you reproduce this issue in IPython?

[fiammante]

It works well with IPython, works also well with docker Tensorflow image with Jupyter. Problem only exists when lauching jupyter from WSL Ubunto .

IPython capture below

(tf) fiammante@DESKTOP-3AQ037Q:~$ ipython
Python 3.9.16 (main, Mar 8 2023, 14:00:05)
Type 'copyright', 'credits' or 'license' for more information
IPython 8.12.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))
2023-05-31 00:40:01.469356: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-05-31 00:40:03.416088: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
2023-05-31 00:40:06.188443: I tensorflow/compiler/xla/stream_executor/cuda/cuda_gpu_executor.cc:982] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node
Your kernel may have been built without NUMA support.
2023-05-31 00:40:06.721602: I tensorflow/compiler/xla/stream_executor/cuda/cuda_gpu_executor.cc:982] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node
Your kernel may have been built without NUMA support.
2023-05-31 00:40:06.721733: I tensorflow/compiler/xla/stream_executor/cuda/cuda_gpu_executor.cc:982] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node
Your kernel may have been built without NUMA support.
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

[mhdpr]

Same problem here on linux

[vaibhavnohria1]

@JasonWeill I would like to work on this issue

[JasonWeill]

@vaibhavnohria1 Thanks for your interest! I've assigned you to this issue; please submit a pull request when your change is ready for review.