machine.dev vs
RunsOn.
RunsOn requires an AWS account. We don't. Spot is 1.4–2.66× cheaper at every CPU size. T4G ARM is the cheapest GPU runner anywhere. No €300/yr licence, no CloudFormation, no AMI maintenance.
Different models.
RunsOn is a fundamentally different product to machine.dev. It's not a SaaS runner service — it's software you install in your own AWS account via CloudFormation. The licence is flat (€300/year), and your compute costs go straight to your AWS bill at raw EC2 rates with no markup.
That's well-engineered and economically attractive at scale. It's also a commitment: an AWS account, an IAM setup, an AMI maintenance discipline, and ongoing service-limit management. RunsOn's on-demand pricing is cheaper than ours at every CPU size — we're not going to bury that. If you're a team with existing AWS expertise running high CI volume, RunsOn deserves serious consideration.
What we offer: spot pricing 1.4–2.66× cheaper than RunsOn spot at every size, the only ARM GPU runner in the market (T4G), pre-built CUDA images, no AWS account, no IAM, no licence fee, no AMI maintenance, and a free tier with no credit card. For most teams that's the right tradeoff. For some teams it isn't — we have a decision matrix below.
Spot wins, OD loses.
Per-minute USD. RunsOn prices verified from runs-on.com/pricing April 9, 2026 in us-east-1, 40 GB gp3 included. RunsOn does not include the €300/yr software licence in these per-minute rates.
[NOTE_HONEST] Read this table top-to-bottom. machine.dev wins decisively on spot at every size. RunsOn wins decisively on on-demand at every size. The structural reason: RunsOn passes raw EC2 rates through to your AWS bill with zero markup; machine.dev runs the typical 2× managed-CI markup on EC2 on-demand.
[NOTE_LICENCE] RunsOn's per-minute prices above don't include the €300/year software licence. At low CI volume (under ~20,000 minutes/year on a 4-vCPU runner) the licence overhead can swing the math back toward machine.dev's pay-as-you-go.
T4 vs T4G ARM.
RunsOn only publishes T4 in their pricing calculator. machine.dev's T4G ARM is the cheapest GPU runner anywhere — and the only ARM GPU runner in the market.
When to pick which.
Honest framing: RunsOn is the right answer for some teams. Here's how to decide.
| IF YOU… | USE |
|---|---|
| Want zero AWS exposure (no account, no IAM, no VPC) | machine.dev |
| Need pre-built CUDA / cuDNN / PyTorch images | machine.dev |
| Want a free tier with no credit card | machine.dev |
| Need ARM GPU (T4G) — only provider in market | machine.dev |
| Want spot pricing on GPU at scale | machine.dev |
| Need AWS Trainium / Inferentia2 | machine.dev |
| Already run AWS at scale, have an SRE team | RunsOn |
| Want lowest possible on-demand CPU cost | RunsOn |
| Need everything in your own VPC / CloudWatch | RunsOn |
| Need H100 / H200 specifically | RunsOn |
Beyond price.
| SPEC | RUNSON | MACHINE_DEV |
|---|---|---|
| Model | Self-hosted (your AWS account) | Fully managed SaaS |
| AWS account required | Yes (CloudFormation install) | No |
| Setup time | ~10 min CloudFormation install + IAM | Install GitHub App, change runs-on label |
| Annual licence | €300/year flat | No (pay-as-you-go credits) |
| Per-minute compute markup | None (raw EC2 pass-through) | ~2× over EC2 (typical managed CI) |
| Spot tier | Yes (native AWS spot) | Yes (1.4-2.66× cheaper than RunsOn spot) |
| CUDA / drivers / cuDNN | No (you build the AMI) | Yes (12.1 + cuDNN 9.2.1, drivers 555.58) |
| GPU SKU coverage | Any EC2 type (T4, A10G, L4, L40S, V100, A100, H100, H200) — only T4 in their pricing calculator | T4G ARM, T4, L4, A10G, L40S, RTX 6000 + Trainium / Inferentia2 |
| ARM GPU runners | No | Yes (T4G — only provider in market) |
| Free trial | 15-day full licence | $10 free compute, no credit card |
| Data transfer fees | AWS egress rates apply to your bill | None (bundled) |
| Software updates | You apply CloudFormation updates | Zero (managed) |
| AWS service limit management | You | machine.dev |
Four strengths.
Cheaper on every on-demand size
RunsOn is essentially raw AWS EC2 pass-through with no per-minute markup. They make their money on the €300/yr flat licence. machine.dev runs the typical 2× managed-CI markup on EC2 on-demand. Result: RunsOn is 53-109% cheaper on x64 on-demand at every size. If you're at high enough CI volume to cover the licence and you want pure AWS rates, RunsOn wins on raw pricing.
Everything in your own VPC
Because it runs in your AWS account, you choose the region, VPC, subnets, IAM policies, instance types, and AMIs. No data leaves your account. Logs go to your CloudWatch. The compliance story is whatever your AWS account already is.
Any EC2 GPU instance
RunsOn can boot any EC2 instance type, which means H100 and H200 are technically available — though they're not published in their pricing calculator and you'll handle the AMI yourself. machine.dev tops out at RTX 6000 (96 GB VRAM) on the public catalog right now.
Flat licence
€300/year regardless of how much you run. A high-volume team and a low-volume team pay the same licence. If you're running thousands of CI minutes per day, the per-minute markup of any SaaS adds up faster than the flat fee.
Six strengths.
Spot is genuinely cheaper at every CPU size
Even though RunsOn is raw EC2 pass-through, machine.dev's spot tier is 1.4-2.66× cheaper than RunsOn's spot at every Linux x64 size from 2 to 64 vCPU, and 1.4-2.5× cheaper on ARM. We pool spot capacity across regions and instance families to find the cheapest pocket; a single-account RunsOn install can't.
Zero AWS exposure
RunsOn requires an AWS account, an IAM role, a CloudFormation install, and ongoing service-limit management (because EC2 has account-level limits on every instance family, including GPU). machine.dev requires installing a GitHub App and changing a runs-on label. There is no AWS account to create, no IAM, no VPC, no AMI to maintain.
Pre-built CUDA images
RunsOn boots a generic Ubuntu AMI; if you want a GPU runner you build the AMI with NVIDIA drivers, CUDA, cuDNN, and your ML framework yourself. machine.dev's GPU runners ship with drivers 555.58, CUDA 12.1, and cuDNN 9.2.1 baked in. torch.cuda.is_available() returns True on the first run.
The only ARM GPU runner anywhere
T4G ARM 8 vCPU spot is $0.00277/min ($0.166/hr) — the cheapest GPU runner on the public internet, and 18% cheaper than RunsOn's T4 spot. RunsOn doesn't offer ARM GPU at any price.
Free trial, no AWS account, no credit card
RunsOn's free tier is a 15-day full licence — but you still need an AWS account, CloudFormation execution permissions, and the ability to run EC2. machine.dev's free tier is $10 of compute with no credit card and no AWS account required.
One bill, no surprise AWS line items
RunsOn compute hits your AWS bill directly — including data transfer, NAT egress, EBS, and the App Runner overhead. machine.dev bundles data transfer and gives you a single credit balance covering CPU and GPU. No mid-month AWS surprise.
Live per-minute rates
Live per-minute rates for all runner types. No AWS bill to estimate - what you see is what you pay.
Get started free.
$10 free compute on signup. No AWS account needed.