Anyone Having Issues With Getting CUDA Running on a P2 Instance


#1

So I had provisioned an EC2 instance for the earlier version of Lesson 1 using Theano and everything worked fine. I cloned the fastai/fastai repo on the same machine and followed all the instructions and ran the notebook. And got False when the notebook ran torch.cuda.is_available()

In the python console I tried running the same function got back True. So Im not entirely sure whats going on here. Is this a PyTorch install issue?


#2

problem solved. It was an issue with the version of CUDA I was running. Im guessing the FastAI env required CUDA 9 and I was running 8. I looked into the nvidia docs and here were the steps i followed
Get the link address to the pkg based on your system architecture

wget link-to-pkg
sudo dpkg -i cuda-repo-ubuntu1604_9.1.85-1_amd64.deb
sudo apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub
sudo apt-get update
sudo apt-get install cuda


(Peter Coghill) #3

great work. Worked for me. Cheers

Google Cloud Compute (where i’m running my VM) seems to install CUDA 7.5 by default. So any other google users, with this error, this is a valid fix.