Cannot access GPU with Tensorflow on MacOS

[che25]

I would like to run Tensorflow on the GPUs of my Mac through RStudio.

library(tensorflow)
library(keras)
tf$config$list_logical_devices('GPU')

The return value is an empty list.
list()

For the CPU the return value is non-empty.

I have successfully installed tensorflow, and have also tried the development version

# install the development version of packages, in case the
# issue is already fixed but not on CRAN yet.
install.packages("pak")
pak::pak(sprintf("rstudio/%s", c("reticulate", "tensorflow", "keras")))
if (is.null(reticulate::virtualenv_starter()))
   reticulate::install_python()
tensorflow::install_tensorflow()

This runs

tensorflow::as_tensor("Hello World")

/Users/cedlich/.virtualenvs/r-tensorflow/lib/python3.9/site-packages/urllib3/init.py:35: NotOpenSSLWarning: urllib3 v2 only supports OpenSSL 1.1.1+, currently the 'ssl' module is compiled with 'LibreSSL 2.8.3'. See: urllib3/urllib3#3020
warnings.warn(
tf.Tensor(b'Hello World', shape=(), dtype=string)

Diagnostics

reticulate::py_config()
tensorflow::tf_config()
reticulate::import("tensorflow")
reticulate::py_last_error()
sessionInfo()

python: /Users/cedlich/.virtualenvs/r-tensorflow/bin/python
libpython: /Library/Developer/CommandLineTools/Library/Frameworks/Python3.framework/Versions/3.9/lib/python3.9/config-3.9-darwin/libpython3.9.dylib
pythonhome: /Users/cedlich/.virtualenvs/r-tensorflow:/Users/cedlich/.virtualenvs/r-tensorflow
version: 3.9.6 (default, Feb 3 2024, 15:58:27) [Clang 15.0.0 (clang-1500.3.9.4)]
numpy: /Users/cedlich/.virtualenvs/r-tensorflow/lib/python3.9/site-packages/numpy
numpy_version: 2.0.2
tensorflow: /Users/cedlich/.virtualenvs/r-tensorflow/lib/python3.9/site-packages/tensorflow

NOTE: Python version was forced by import("tensorflow")

tensorflow::tf_config()
TensorFlow v2.18.1 (/.virtualenvs/r-tensorflow/lib/python3.9/site-packages/tensorflow)
Python v3.9 (/.virtualenvs/r-tensorflow/bin/python)
reticulate::import("tensorflow")
Module(tensorflow)
reticulate::py_last_error()
NULL
sessionInfo()
R version 4.4.3 (2025-02-28)
Platform: aarch64-apple-darwin20
Running under: macOS Sonoma 14.2.1

Matrix products: default
BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/lib/libRlapack.dylib; LAPACK version 3.12.0

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

time zone: Europe/Berlin
tzcode source: internal

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] keras_2.15.0 tensorflow_2.16.0.9000

loaded via a namespace (and not attached):
[1] R6_2.6.1 base64enc_0.1-3 Matrix_1.7-2 lattice_0.22-6 reticulate_1.41.0.9001
[6] magrittr_2.0.3 rappdirs_0.3.3 generics_0.1.3 png_0.1-8 lifecycle_1.0.4
[11] ps_1.9.0 cli_3.6.4 processx_3.8.6 pak_0.8.0.1 grid_4.4.3
[16] zeallot_0.1.0 withr_3.0.2 tfruns_1.5.3 compiler_4.4.3 rprojroot_2.0.4
[21] here_1.0.1 rstudioapi_0.17.1 tools_4.4.3 whisker_0.4.1 Rcpp_1.0.14
[26] rlang_1.1.5 jsonlite_1.9.1

[t-kalinowski]

To use a GPU on macOS, you also need to install the tensorflow-metal package. We don't install it by default because tensorflow-metal does not fully support all of tensorflow, and just having tensorflow-metal installed causes some tensorflow calls to throw exceptions.

That said, for the well-trodden parts of the API, tensorflow-metal works without any issues.