Tacotron-2: seems gpu not been used

[pandaGst]

I run python3 train.py --model='Tacotron' with default parameters in hparams.py. But when I check GPU using with nvidia-smi, Gpu-Util is 0%.
Is there any one got same question?
Thank you very much!

[sid-sundrani]

Check out how to find the connected gpu devices on tensorflow- https://stackoverflow.com/questions/38009682/how-to-tell-if-tensorflow-is-using-gpu-acceleration-from-inside-python-shell.

If the gpu device is not in there, make sure you have tensorflow-gpu installed

[pandaGst]

@sid-sundrani Thank you very much!
This issue troubled me few days.
I will try as your suggestion!

[pandaGst]

I changed some package's verson of my environment:
tf-gpu==1.10
cuda==9.0
cudnn==7.1
anaconda==3
python==3.6.7

other packages version is same as requirment.txt gave.

Then it worked.
But the capacity of gpu is not hight, just about 20%.

[iamjanvijay]

Thanks. @pandaGst It worked for me as well now. I agree GPU utilization is really low.