I am trying to train an object detection model on a laptop, which does not have Nvidia GPU, so have to use tensorflow-directml-plugin instead.
I am following the official instructions from MS as found here: https://learn.microsoft.com/en-us/windows/ai/directml/gpu-tensorflow-plugin
Here are the TF packages that I have in my venv: tensorflow-addons 0.21.0 tensorflow-cpu 2.10.0 tensorflow-datasets 4.9.0 tensorflow-directml-plugin 0.4.0.dev230202 tensorflow-estimator 2.10.0 tensorflow-gpu 2.10.0 tensorflow-hub 0.14.0 tensorflow_intel 2.10.0 tensorflow-io 0.31.0 tensorflow-io-gcs-filesystem 0.31.0 tensorflow-metadata 1.13.0 tensorflow-model-optimization 0.7.5 tensorflow-object-detection-api 0.1.1 tensorflow-text 2.10.0
python = 3.10.2 TF<=2.10.* (should support GPU)
Machine: CPU = i7-8665U (appears on the list of supported devices) GPU = Intel UHD Graphics 620 (>500 -> should work fine). GPU driver v31.0.101.* (the latest) OS = Win 10 19045.3208 (also meeting requirements here)
print(tf.config.list_physical_devices('GPU')) confirms that 1 GPU is connected. [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
However, when I am training a model, I am also not seeing any improvements in computational power. The script runs fine without any errors and confirms that it is attempting to engage GPU, but GPU load is at 6%, which is the same as the level I have when idling. Any ideas on why it could be the case?
