Tensorflow GPU on ARM M1 Pro

[SimoneBrazzi]

Good morning,

I have followed the installation guide for tensorflow fro R. I have a M1 Pro, so in my miniconda env I have installed tf and the metal extension, following the apple guide

Tensorflow Plugin - Metal - Apple Developer

This means I can use tf with my GPU. here is my doubt: what does the following message means?

library(tensorflow)
tf$constant("Hello tf!")
2023-08-31 09:09:38.772047: I metal_plugin/src/device/metal_device.cc:1154] Metal device set to: Apple M1 Pro
2023-08-31 09:09:38.772092: I metal_plugin/src/device/metal_device.cc:296] systemMemory: 16.00 GB
2023-08-31 09:09:38.772106: I metal_plugin/src/device/metal_device.cc:313] maxCacheSize: 5.33 GB
2023-08-31 09:09:38.772254: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:303] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2023-08-31 09:09:38.772633: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:269] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: )
tf.Tensor(b'Hello tf!', shape=(), dtype=string)

tensor flow::tf_config()
TensorFlow v2.13.0 (~/miniconda3/envs/myenv/lib/python3.11/site-packages/tensorflow)
Python v3.11 (~/miniconda3/envs/myenv/bin/python3.11)

I will try to check using my activity monitor if it is working on my cpu or gpu. Is there a way to suppress this message?

Thanks,

Simone

[technocrat]

I installed with /usr/bin/python3 and put it at the top of my $PATH, so I'd guess I'd blame Anaconda.

[SimoneBrazzi]

Thank you for answering me, but I don't get it: do you mean if I avoid using miniconda it should work?

Thanks

[technocrat]

No guarantees, but there have been many reports here of conda related install problems of all kinds. It’s possible that those might be fixable with enough tinkering with PATH variables and such, but using this might be the easier way to go.