Torch.cuda.is_available() returns false while torch.backends.cudnn.enabled returns true

(Divyansh Rai) #1

I;m using google colab and I have already checked and the hardware accelerator is set to GPU, but even so on my clouderizer when I try to run the torch.cuda.is_available() it returns false. I also changing the machine

I also tried changing the machine settings and checking on install CUDA 8.0, but it doesn’t change anything.

(Cedric Chee) #2

Looks like your NVIDIA driver install is corrupted. Check your NVIDIA driver. Does nvidia-smi look fine?


  1. change the machine to use CPU, wait for a few minutes, then change back to use GPU
  2. reinstall the GPU driver

(Divyansh Rai) #3

Turns out, I had to uncheck the CUDA 8.0