Alternative for "!nvidia-smi" for Apple Silicon based GPU

[Mathews-Tom]

Hi Daniel

I am using an Apple Silicon M1 Pro Macbook. I am currently working with Tensorflow v2.8.0 with TensorFlow-Metal for Apple Silicon. For the course work, I am using Google Colab running on a local Jupyter Notebook server. Working through different deep learning coding and running TensorFlow I do see that log messages, especially while fitting a model, that I am using GPU through TensorFlow-Metal.

On a few of the notebooks, I have noticed that you have used the following command in the notebook.

# Are we using a GPU?
!nvidia-smi

I am guessing this command is specifically meant for Intel based GPUs. Can you please let me know if there is an alternative for Apple M1 based GPU?

[mrdbourke]

Hey @Mathews-Tom,

Yes you're right !nvidia-smi is specifically for NVIDIA GPUs.

As for Mac there's no real dedicated way but you can try: https://www.tensorflow.org/api_docs/python/tf/config/list_physical_devices

Run this code in a Jupyter cell/Python environment with TensorFlow available:

import tensorflow as tf

physical_devices = tf.config.list_physical_devices('GPU')
print(physical_devices)

The output should be a list of the devices available to TensorFlow.

[Mathews-Tom]

Thanks for the feedback. @mrdbourke

[gauravyadav016-png]

As an alternative to nvidia-smi is Activity Monitor. Search on spotlight (cmd + space) and type Activity Monitor. When the program opens press cmd + 4. This will show you the active usage of GPU(s) on your system