How to Set Up GPU GitHub Larger Runners

GPU larger runners are a specific size you select under the GPU-powered option, useful for machine-learning and rendering jobs, and they bill at a higher per-minute rate.

According to GitHub's docs, GitHub-owned GPU larger runners are available, with a documented configuration of 4 vCPU and 1 GPU (Tesla T4), 28 GB RAM, 16 GB VRAM, and 176 GB storage for Ubuntu and Windows. GPU runners bill per minute like other larger runners. Sources: https://docs.github.com/en/actions/reference/runners/larger-runners, https://docs.github.com/en/actions/using-github-hosted-runners/using-larger-runners/about-larger-runners, https://docs.github.com/en/billing/reference/actions-runner-pricing.

Create the GPU runner

In Settings, Actions, Runners, create a new GitHub-hosted runner.
Select a size under the GPU-powered option (per GitHub docs); base Windows 11 desktop, GPU-optimized, and ARM64 images appear under the partner tab.
Assign a runner group and a clear name, for example ubuntu-gpu-t4.

Target it and confirm the GPU

.github/workflows/train.yml

jobs:
  train:
    runs-on:
      labels: ubuntu-gpu-t4
    steps:
      - uses: actions/checkout@v4
      - run: nvidia-smi   # verify the GPU is visible
      - run: ./train.sh

Mind the per-minute cost

According to GitHub's pricing reference (retrieved 2026-07-02), a 4-core GPU Linux larger runner was listed at $0.052/min, higher than a same-core standard runner. GPU jobs are worth it when they cut wall-clock time enough to offset the rate. Verify current pricing at https://docs.github.com/en/billing/reference/actions-runner-pricing.

When to consider a managed option

GPU pipelines often mix long training steps with flaky setup steps. Latchkey managed runners add self-healing so a transient failure early in a GPU job retries automatically instead of wasting the whole billed run. Evaluate it alongside first-party GPU runners for your workload.