GPU-util is 0% on YOLOv3 training

Hi! I’m trying to train a YOLOv3. I have RTX 3060Ti. I can performance all the training but the use of GPU is ~0%. The source code of my model, training and dataset logic is here

I found some reference to change the load of items, but I can’t understand entirely how to make the changes. Thanks for your help :smiley:

I don’t think your current outputs show the wanted utilization and you should be able to select a “Compute” tab in task manager (I’m not deeply familiar with Windows). Alternatively, use nvidia-smi to check the GPU util.

1 Like

Yes, I search more options and found that Task Manager cna show CUDA usage
image