from Hacker News

Ask HN: How do I run PyTorch with GPU on a windows machine with AMD Radeon?

by brauhaus on 7/9/23, 1:45 PM with 1 comments

I feel I'm ready to leave the Colab environment, and I wanted to start developing PyTorch projects on my local machine. The problem: I have a Windows machine with AMD Radeon GPU (cries in wsl)

What is the preferred way to get started? I've tried using WSL, ROCm, and PyTorch-DirectML, but I'm always running into problems when I run `to(device)`.

  • by vkaku on 7/9/23, 3:35 PM

    Look at Microsoft's Antares. They use a wierd combination of LoadLibrary and WSL to run ROCM workloads.

    TL;DR not a good use of your time if AMD doesn't want to ship ROCM on Windows for all customers.

    Read this if you want to do something like it: https://github.com/microsoft/antares/blob/v0.3.x/backends/c-...

    Source: been on this path before and moved on.