Testing GPU code

[bwehlin]

Why are you starting this discussion?

Question

What GitHub Actions topic or product is this about?

Actions Runner

Discussion Details

Hi,

I'm developing an open-source Python module with CUDA support. I'm using GitHub Actions as my CI, which I'm liking very much!

There is an obvious issue that I can't test any of my GPU code as part of my Actions run as none of the runners available to open-source projects have GPU support (as far as I can tell).

Is there any possibility that GPU support could be added to the small runners in the future? For me, there is no requirement that a particularly big GPU is available. I just want to unit test some small computations (likely less than ten seconds of GPU time per test suite run).

Thanks!

[MasteraSnackin]

Hey, yeah unfortunately the GitHub-hosted runners don't have GPU support for open-source projects right now — you're not missing anything, that's just how it is at the moment.

The workaround most people use is setting up a self-hosted runner on a machine (or a cloud VM) that has a GPU and CUDA installed, then targeting it with a label in your workflow:

jobs:
  gpu-tests:
    runs-on: [self-hosted, linux, gpu]
    steps:
      - uses: actions/checkout@v4
      - name: Run GPU tests
        run: pytest tests/

For cloud, a single T4 on something like Lambda Labs or Paperspace is pretty affordable if you only need a few seconds of GPU time per run. You could also just split your test suite — keep the CPU tests on the default hosted runners and only route the CUDA-specific ones to the self-hosted runner. That way you're not paying for GPU time on tests that don't need it.

As for whether GitHub will ever add even a small GPU to the free hosted runners, I haven't seen anything on the roadmap for it, but it would genuinely be a useful feature given how common CUDA projects are these days. Hopefully it gets picked up eventually.