I'd like to do some machine learning on my AMD 6800 XT gpu within a python image based on python:3.9.10. I can confirm that the GPU is available outside of the image (in a wsl2 instance). However, if I do docker run -it python:3.9.10 /bin/bash and then complete the same tutorial (https://docs.microsoft.com/en-us/windows/ai/directml/gpu-tensorflow-wsl#install-the-tensorflow-with-directml-package) it doesn't work:

(directml) root@8a8274e5337f:/# python
Python 3.6.13 |Anaconda, Inc.| (default, Jun  4 2021, 14:25:59)
[GCC 7.5.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow.compat.v1 as tf
>>> tf.enable_eager_execution(tf.ConfigProto(log_device_placement=True))
>>> print(tf.add([1.0, 2.0], [3.0, 4.0]))
2022-08-18 12:29:39.540717: W tensorflow/stream_executor/platform/default/dso_loader.cc:108] Could not load dynamic library 'libdirectml.0de2b4431c6572ee74152a7ee0cd3fb1534e4a95.so'; dlerror: libd3d12.so: cannot open shared object file: No such file or directory
2022-08-18 12:29:39.540760: W tensorflow/core/common_runtime/dml/dml_device_cache.cc:137] Could not load DirectML.
2022-08-18 12:29:39.540793: I tensorflow/core/common_runtime/dml/dml_device_cache.cc:250] DirectML device enumeration: found 0 compatible adapters.
2022-08-18 12:29:39.541010: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2022-08-18 12:29:39.545145: I tensorflow/core/common_runtime/eager/execute.cc:571] Executing op Add in device /job:localhost/replica:0/task:0/device:CPU:0
tf.Tensor([4. 6.], shape=(2,), dtype=float32)

This article has led me to think that perhaps docker doesn't support AMD GPUs at all: https://docs.docker.com/desktop/windows/wsl/

Can anyone suggest what I might be able to do to get this to work?

Note that the reason I have picked this image is that my environment is based on a rather lengthy Dockerfile inheriting from python:3.9.10, and I'd like to keep using that image on the PC with the GPU as well as other (nvidia) environments, so I'm after a portable solution as far as the image is concerned, although I'd be grateful for any solution at this point.

  • Basically you are just passing it in with the --device flag but you need drivers and stuff. Check this project out github.com/RadeonOpenCompute/ROCm-docker the details section is really helpful.
    – Levi
    Aug 18, 2022 at 13:49

1 Answer 1


Tensorflow doesn't have access to libd3d12.so file:

2022-08-18 12:29:39.540717: >tensorflow/stream_executor/platform/default/dso_loader.cc:108] Could not load dynamic library 'libdirectml.0de2b4431c6572ee74152a7ee0cd3fb1534e4a95.so'; dlerror: libd3d12.so: cannot open shared object file: No such file or directory

Here is an article solving this issue: enter link description here

Shortly said, use this line instead:

docker run -it --rm --device /dev/dxg --mount type=bind,src=/usr/lib/wsl,dst=/usr/lib/wsl -e LD_LIBRARY_PATH=/usr/lib/wsl/lib python:3.9.10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.