Get an error when using DirectML

[mariosconsta]

I have an AMD GPU and I came upon this tool so I thought I could try it out! I did pip install tensorflow-directml, restarted my kernel and now I get this error when I run this code:

a = tf.constant([1.0, 2.0, 3.0]) 

b = tf.constant([4.0, 5.0, 6.0]) 

c = tf.add(a, b)

print(c)

InvalidArgumentError: Multiple OpKernel registrations match NodeDef at the same priority '{{node AddV2}}': 'op: "AddV2" device_type: "GPU" constraint { name: "T" allowed_values { list { type: DT_FLOAT } } }' and 'op: "AddV2" device_type: "GPU" constraint { name: "T" allowed_values { list { type: DT_FLOAT } } }' [Op:AddV2]

Can I not use my GPU after all?

[PatriceVignola]

Hi @mariosconsta , did you mean tensorflow-directml-plugin instead of tensorflow-directml? If yes, this is already a known issue (#216). The solution at the moment is to use the tensorflow-cpu plugin together with tensorflow-directml-plugin, NOT tensorflow or tensorflow-gpu. We have a PR open against the TensorFlow Core repo that hopefully will get merge soon (tensorflow/tensorflow#55645), so use tensorflow-cpu in the meantime.

Functionally speaking, all packages are equivalent when using pluggable devices so your GPU will still be used, even though the package is named tensorflow-cpu.