Cutting costs for GPU-powered GitHub runners

Published on October 4, 2024 | Written by Andreas

GPU-powered GitHub runners have emerged as a game-changing solution, offering developers the ability to harness the immense processing power of graphics processing units for tasks such as machine learning, data analysis, and complex simulations. Recognizing this need, GitHub announced GPU-powered runners in July 2024, bringing advanced computational capabilities directly to its platform. However, for teams looking for a more budget-friendly option, HyperEnv has stepped up to provide a cost-effective alternative, offering GPU-powered runners with potential cost savings of up to 75%.

HyperEnv provides a convenient way to run self-hosted GitHub runners on AWS, which enables you to reduce costs, access private networks, and customize the build environment.

Cutting costs for GPU-powered GitHub-hosted runners

GitHub charges $0.07 per minute for a GPU-powered larger runner (Linux, 4 CPU cores). Also, GPU-powered runners are only available for GitHub Team or GitHub Enterprise Cloud plans.

The following cost estimation shows what it costs to run 5000 build minutes on an g4dn.xlarge instance managed by HyperEnv.

So by switching from GitHub-hosted runners with GPU to HyperEnv you are able to cut costs by about 75%. Go to Pricing to run the cost estimation by entering your own numbers.

Next, you will learn how to run a GitHub job with access to an NVIDIA T4 GPU on AWS by using HyperEnv.

Run GitHub workflow on AWS with access to NVIDIA T4 GPU

First, get started with HyperEnv by following the setup guide. During the process select the instance type g4dn.xlarge (4 CPU, 16 GB memory, NVIDIA T4 GPU).

Second, change the runs-on section of the GitHub job that requires a GPU to ['hyperenv'].

---
name: 'gpu-runner'
on:
  workflow_dispatch:
jobs:
  nvidia:
    runs-on: ['hyperenv'] # Run job on HyperEnv
    steps:
    - uses: 'actions/checkout@v4'
    - name: 'NVIDIA SMI output'
      run: |
                nvidia-smi

When a GitHub job is started, HyperEnv will spin up an EC2 instance of type g4dn.xlarge providing 4 CPU cores, 16 GB memory, and an NVIDIA T4 GPU.

Summary

Are you running jobs on GPU-powered runners provided by GitHub? Consider switching to HyperEnv to cut costs by up to 75%. Get started with HyperEnv today!.