Hello, I am learning tensorflow. I am having a very annoying problem with trying to run tensorflow on my AMD GPU. I have set up a conda environment running tensorflow 2.9, python 3.9.12, with tensorflow-directml-plugin because I don’t have an nvidia GPU. Whenever I try to use tensorflow for basically anything, this message appears. I cannot find an answer online that I understand. If somebody has any idea what is going on, or at least point me towards something, it would be very helpful
Unfortunately, you cannot use tensorflow or tensorflow-gpu yet with this plugin because the “GPU” device type conflicts with the built-in GPU type for CUDA devices. This is something that will hopefully get fixed in TensorFlow 2.10 (we already proposed a change that should address it: Fix duplicate kernel registration errors when using a Pluggable Device with the `GPU` type by PatriceVignola · Pull Request #56707 · tensorflow/tensorflow · GitHub)
In the meantime, please use the tensorflow-cpu package when using tensorflow-directml-plugin.
2 Likes