Using 0% of GPU

[kast1450]

It's purely using my CPU.
I can't find anything about GPU settings other than multi_gpu, which spits out an error saying "Division by zero."

Is there anything I'm missing?

Using version 1.4.1 straight from Pip, and I have tensorflow-gpu installed.

[minimaxir]

Are you able to use the GPU with other TensorFlow-based frameworks? (i.e. you're using an nVidia GPU w/ CUDA installed correctly)

[kast1450]

I'm using a GTX 1060 and I have the drivers installed, do I need to install the CUDA Toolkit (2.1GB)?

[minimaxir]

Likely yes, with CuDNN as well. (I'm not an expert on setting up CUDA/CuDNN on local computers, but there are tons of Deep Learning tutorials on the internet saying how)

[kast1450]

Not sure what I'm supposed to do with the CuDNN library as it seems to be for C++ development.

Installing the toolkit just says that it will install an older version of the Nvidia drivers even thought I got the latest CUDA Toolkit available.

Uninstalling tensorflow and only using tensorflow-gpu just spits out an error about tensorflow missing.

[frnsys]

I'm having this same issue -- other scripts using tensorflow utilize the GPU correctly, but for some reason it stays at 0% with this library.

Edit: tensorflow does see the GPU when I run textgenrnn:

Created TensorFlow device 
(/job:localhost/replica:0/task:0/device:GPU:0 with 10409 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:02:00.0, compute capability: 6.1)

but doesn't use it for training.

[oparkerj]

I have this same issue. I installed tensorflow-gpu, as well as CUDA and cuDNN but still nothing.

[kast1450]

I have this same issue. I installed tensorflow-gpu, as well as CUDA and cuDNN but still nothing.

Hey! I fixed this by installing TensorFlow-gpu version 1.6.0 by using pip. (pip install tensorflow-gpu==1.6.0) then installing the Cuda Toolkit v9.0.0, and CuDNN v7.0.5 for Cuda 9.0.

now it trains on GPU just fine! hope this helps

[oparkerj]

Thank you! I'm not sure which specific one it was, but using the versions you mention solved the issue.

[kast1450]

Thank you! I'm not sure which specific one it was, but using the versions you mention solved the issue.

No problem man. Glad I could help.

[minimaxir]

Good to hear the issues were resolved! :)