Each time I run the fit method for keras model, I got "...loss: nan - val_loss: nan"

System configuration: windows 10 tensorflow 1.13.1 Keras 2.2.4

GPU settings: Type: NVIDIA GeForce 940MX Driver version: cuda toolkit file: cuda_10.0.130_411.31_win10.exe CUPTI file: cupti-win64-cuda10-1.0.7.zip cuDNN SDK file: cudnn-9.0-windows10-x64-v7.1.zip

Error details: Failed to initialize GPU device #0: unknown error


I was able to get around this by changing the way the CUDA orders the devices. I'm not entirely sure why this worked. For python, you can do this with:

import os

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"

I am using an mx150(1D10), which is the 10 series equivalent of the 940mx listed in your screenshot

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.