machine.dev
GPU, x64 and arm64 runners for GitHub Actions on dedicated AWS instances, with spot or on-demand tenancy chosen in the runs-on label. A GPU-first managed runner service for GitHub Actions built on AWS instances, suited to machine-learning teams and to anyone wanting spot-priced CPU runners.
Overview
machine.dev provides GPU and CPU runners for GitHub Actions. After installing its GitHub App, a job selects hardware in the label itself, for example runs-on: machine/gpu=T4/tenancy=spot or machine/cpu=8/architecture=arm64. Each job gets a dedicated AWS instance running Ubuntu 22.04 that registers as an ephemeral self-hosted runner and shuts down afterwards. GPU images ship with NVIDIA drivers, CUDA and the NVIDIA Container Toolkit installed.
Six NVIDIA GPU tiers, AWS Trainium and Inferentia2 accelerators, and x64 or Graviton CPU runners from 2 to 64 vCPU are offered, each as spot or on-demand, across six AWS regions. Billing is per minute of runtime in US dollars, with EBS storage charged separately. The pricing reference gives “from” rates because spot prices follow the AWS market, and concurrency depends on the plan.
Notable
- 2024: company founded by Leonard O’Sullivan and Lyndon Swan, described as remote-first (about page).
- June 20, 2025: the post “Why We Built machine.dev” was published.
- April 27, 2026: nat-zero, a Terraform module for scale-to-zero NAT instances on AWS, was open sourced.
- June 12, 2026: per-job instance metrics became generally available, on by default.
machine.dev specifications
| Attribute | Details |
|---|---|
| CI systems | GitHub Actions |
| Operating systems | Linux |
| Architectures | x64, arm64 |
| Deployment | Managed runner service |
| Isolation | Each job runs on a dedicated AWS instance that registers as an ephemeral self-hosted runner and shuts down when the job finishes |
| Fresh environment per job | Yes |
| Hardware | AWS instances: x64 (Intel/AMD) and arm64 (Graviton) CPU runners, each with a default 100 GB gp3 EBS volume (6,000 IOPS, 250 MB/s) that can be raised to 16 TB. |
| Performance (vendor claim) | machine.dev says a runner goes from push to running job in under 60 seconds without pre-warming (vendor claim) |
| Runner sizes | CPU runners from 2 vCPU / 4 GB to 64 vCPU / 128 GB on x64 and arm64; GPU runners from 4 vCPU / 8 GB (T4G) to 192 vCPU / 2,048 GB with 8 RTX 6000 |
| Caching | An Actions cache served from the same AWS region as the runner is in closed beta, available by request: 25 GiB per repository per region with least-recently-used eviction, not billed during the beta. It works with actions/cache v4+ and BuildKit type=gha without workflow changes. |
| Docker layer caching | Not documented |
| GPU | NVIDIA T4G, T4, L4, A10G, L40S and RTX 6000 (up to 8 per runner) plus AWS Trainium and Inferentia2; a 4-vCPU T4G runner is listed from $0.00358/min on spot and $0.01400/min on-demand. |
| Concurrency | Pay-as-you-go: 1 concurrent runner before a purchase and 5 after; Builder 10, Developer 25, Growth 30, Pro 40, Business 50. Jobs above the cap wait in arrival order for up to six hours |
| Job limits | 30 minutes per job for non-paying teams; paying teams can run up to GitHub’s 5-day self-hosted runner limit. A spot job interrupted while running is not re-run or resumed automatically |
| Static IPs | Not documented |
| SSH into running jobs | Not documented |
| Observability | Per-job CPU, memory, disk, network and GPU metrics on the dashboard, plus a cost breakdown for every job |
| Setup | Sign up, install the Machine Provisioner GitHub App on an account or organization, enable self-hosted runners for the repositories, then set runs-on to a machine/... label. |
| Example label | machine/gpu=T4/tenancy=spot |
| Pricing | Per-minute USD “from” rates that vary by region. x64 2 vCPU / 4 GB: from $0.00076 spot or $0.00297 on-demand; arm64 2 vCPU / 4 GB: from $0.00040 spot or $0.00241 on-demand; x64 64 vCPU: from $0.01320 spot or $0.09520 on-demand. EBS storage is billed separately at $0.08 per GB-month, prorated to runtime; optional monthly plans from $25 to $300 add bonus credit and higher concurrency. |
| Free tier | $10 of free compute on signup; before any purchase, one job at a time and 30 minutes per job |
| Compliance | Not documented |
| Regions | Six AWS regions: us-east-1, us-east-2, us-west-2, ca-central-1, eu-south-2 and ap-southeast-2, selectable with a regions= option in the label; GPU availability varies by type |
| Open source | No |
| Company headquarters | Queensland, Australia · More from Australia |
Facts from machine.dev’s public documentation, verified November 22, 2026. Performance figures are vendor claims.
Strengths and trade-offs
Reasons to choose it
- You need NVIDIA GPUs in GitHub Actions jobs, from a T4G up to 8 RTX 6000 on one runner
- You want to choose spot or on-demand tenancy per job in the runs-on label
- You need to pin jobs to specific AWS regions, such as eu-south-2 for EU data residency
- You want x64 or Graviton arm64 CPU runners up to 64 vCPU with a configurable EBS volume
Trade-offs
- A spot job interrupted while running is not re-run or resumed automatically
- Runners share outbound IP addresses with other jobs (FAQ)
- The Actions cache is in closed beta and available by request
- All runners use Ubuntu 22.04 LTS; Windows and macOS runners are not documented
Good fit for: Teams on GitHub Actions that run GPU workloads such as model tests or fine-tuning in CI, or want spot-priced x64 and arm64 runners on AWS.
Use cases machine.dev fits
- Lower CI cost: Spot tenancy is listed from $0.00076 per minute for a 2-vCPU x64 runner, with on-demand as a fallback option.
- ARM64 builds: Graviton arm64 CPU runners from 2 to 64 vCPU are selected with an architecture=arm64 option.
- GPU jobs: Six NVIDIA GPU tiers and AWS Trainium and Inferentia2 runners are listed with CUDA and drivers preinstalled.
How machine.dev compares
| Runner | CI systems | Runs on | Deployment | Isolation | Price | Largest runner | Docker layer cache | GPU | Open source | Company HQ |
|---|---|---|---|---|---|---|---|---|---|---|
| machine.dev | GitHub Actions | Linux, x64, arm64 | Vendor-hosted | Dedicated AWS instance | $0.00297/min on-demand | 64 vCPU / 128 GB | Not documented | Yes | No | Australia |
| Cirun | GitHub Actions | Linux, macOS, Windows, x64, arm64 | Your cloud or hardware | Cloud VM | $29/mo (3 private repos) + cloud cost | Set by your infrastructure | Yes | Yes | No | United Kingdom |
| RunsOn | GitHub Actions | Linux, Windows, x64, arm64 | Your cloud or hardware | EC2 instance | €300/yr licence + AWS cost | 896 vCPU | Yes | Yes | No | France |
| Namespace | GitHub Actions, Buildkite | Linux, macOS, Windows, x64, arm64 | Vendor-hosted | Virtual machine | $0.002/min prepaid, $0.003/min overage (2 vCPU / 4 GB) | 64 vCPU / 512 GB | Yes | Not documented | No | Not documented |
| Ubicloud Runners | GitHub Actions | Linux, x64, arm64 | Vendor-hosted | KVM VM | $0.00125/min | 30 vCPU / 120 GB | Yes | Not documented | Yes | United States |
| WarpBuild | GitHub Actions | Linux, macOS, Windows, x64, arm64 | Vendor-hosted | Ephemeral VM | $0.004/min | 32 vCPU / 128 GB | Yes | Not documented | No | Not documented |
| Blacksmith | GitHub Actions | Linux, macOS, Windows, x64, arm64 | Vendor-hosted | Firecracker microVM | $0.004/min | 32 vCPU / 128 GB | Yes | Not documented | No | United States |
| GitHub-hosted runners | GitHub Actions | Linux, macOS, Windows, x64, arm64 | Platform-hosted | Fresh VM | $0.006/min | 96 vCPU / 384 GB | Not documented | Yes | No | United States |
Head-to-head: Cirun vs machine.dev
Frequently asked questions
What is machine.dev?
Which CI systems does machine.dev work with?
How does machine.dev isolate jobs?
Is machine.dev open source?
How much does machine.dev cost?
How do I switch from GitHub-hosted runners to machine.dev?
machine/gpu=T4/tenancy=spot.How fast is machine.dev?
Does machine.dev have a free tier?
Does machine.dev support ARM64 runners?
Does machine.dev offer GPU runners?
How many jobs can run at once on machine.dev?
Where do machine.dev runners run?
Where is MACHINE DOT DEV TRADING PTY LTD based?
What are the alternatives to machine.dev?
Sources
- machine.dev/
- machine.dev/pricing/
- machine.dev/docs/
- machine.dev/docs/reference/pricing/
- machine.dev/docs/platform-specifications/cpu-runners/
- machine.dev/docs/platform-specifications/regions/
- machine.dev/docs/reference/faq/
- machine.dev/docs/guides/cache/
- machine.dev/gpu-runners/
- machine.dev/cpu-runners/
- machine.dev/about-us/
- machine.dev/terms-of-service/
- machine.dev/blog/
Check before you rely on it. Products change quickly, and this page reflects public documentation as of November 22, 2026. Confirm anything that matters in the provider’s own documentation.
Cite this page
CI Runners. (2026). machine.dev. cirunners.org. Last reviewed November 22, 2026. https://cirunners.org/runners/machine/