GPU utilization is low during training

Hi everyone :grinning:
I’m having a problem on my laptop, where the task manager shows the GPU utilization is fairly low (mostly below 38%) during training, but the power consumption is very high. I’m fairly new to pytorch and programming with GPU and use it mostly for academic assignments. Is it possible to improve the utilization of the GPU to make the training more efficient?

I’ll try to provide as much info as I can:
OS: windows 10
CPU: Intel(R) Core™ i7-7500U
RAM: 8.00 GB
GPU: Nvidia Geforece GTX 950M

I’m working on Pycharm pro and sometimes Jupyter Notebook on chrome.
During training the task manager shows the GPU is < 38%, Memory is ~83% and CPU ~34%.
I’ve checked that my GPU is available, and during training I see my .py file on the Nvidia control panel.

I would recommend to check the Performance Guide and check if you could enable e.g. cudnn.benchmark or use automatic mixed-precision training.
Also, you could profile the code using the PyTorch Profiler and see where the bottlenecks are (e.g. in the data loading or the model itself).