Google Colaboratory local runtime using local GPU

Question

I'm using Colaboratory and Pytorch to run a GAN that uses an unusual dataset, which is currently stored locally on my machine. To access these files I connected to a local runtime (as per https://research.google.com/colaboratory/local-runtimes.html). However, Colaboratory now uses my own GPU when running now, which it did not do on previous runs. I know this because current runs run much much slower, as they are using my GTX 1060 6GB instead of Colab's Tesla K80.

I checked this using

torch.cuda.get_device_name(0)

Which returns "GeForce GTX 1060 6G" when I am connected locally. This is the case even with Edit -> Notebook Settings -> Hardware Accelerator -> "GPU" selected.

However, when I am not connected locally, and instead use the (default) "Connect to hosted runtime" option,

torch.cuda.get_device_name(0) does return "Tesla K80".

I've had trouble uploading my dataset to Drive, as it is a large image dataset, and would like to carry on using the local runtime.

How do I use both the local runtime and Colab's amazing Tesla K80? Any help would be much appreciated.

Answer 1

Colab is using your GPU because you connected it to a local runtime. That's what connecting it to your own runtime means. It means that you're using your machine instead of handling the process on Google's servers. If you want to still use Google's servers and processing capabilities, I'd suggest looking into connecting your Google Drive to the Colaboratory runtime.