Low GPU usage during training

General Discussion
,
#1

I’m using Tensorflow 2.10 on Windows 10 and GPU usage typically stays pretty low during training.

I’m running the official RNN classification tutorial from Classification de texte avec un RNN  |  Text  |  TensorFlow basically as it is and during the training the GPU usage (as reported by task manager) hovers around 10%. And the training subsequently takes too long. My GPU is a GeForce RTX 2080 Super with 8GB GPU Memory (system memory is 64GB) so it’s not the most advanced GPU but still I’m not sure why the GPU usage is so low and what I can do to improve it. The GPU memory seems to be utilized well though. It hovers around 7.5GB through out training. Not sure if that’s somehow affecting GPU usage? CPU usage is pretty low as well so no bottleneck there. Any insights or additional troubleshooting steps would be appreciated.

#3

Try to increase the Batch Size of the training data and see if it makes any difference. Thank you!

#5

Thank you. That definitely helped.