I thought I’d share this super cool finding with this community.
For a while now, if you had a Windows PC, you could use a locally installed NVIDIA GPU for deep learning tasks as cuda drivers are now available for windows.
However, this was suboptimal in a few ways. Most code is tested and developed on linux builds, and in some cases python libraries operate differently in windows than in OSX. I have run into this multiple times with the multiprocessing library for example. This has meant that it’s usually still easier to develop deep learning on a linux build if you can.
However, now, for windows PC users with NVIDIA graphics cards installed, there is now an even better solution! You can run windows, start Ubuntu as a virtual machine inside your Windows environment, and… make the Ubuntu virtual machine use your NVIDIA GPU for deep learning! Incredible!
The instructions are here: https://docs.nvidia.com/cuda/wsl-user-guide/index.html
The only part that really takes a while is making sure you are running Windows 10 with the special developer build 20145 or higher. You need to make sure you are using WSL 2, and upgrade the kernel after upgrading windows. Other than that, its pretty straightforward!
I am officially deprecating my Ubuntu boot drive I used to use for. deep learning tasks.