SKIP_TO_MAIN_CONTENT
06.0 // Comparison Apples to apples Same YAML

machine.dev vs
AWS CodeBuild.

CodeBuild's gpu1.large costs $0.30/min — that's $18/hr for a single GPU, the highest published GPU rate in the market. machine.dev's T4 4-vCPU spot runner is $0.00451/min: 66× cheaper.

01 Intro
[REF_001]

First-party. Most expensive.

AWS CodeBuild Managed GHA Runners is AWS's first-party way to run GitHub Actions inside your AWS account. You add a CodeBuild project, an IAM role, a buildspec, and switch your runs-on to codebuild-<project>. The integration with IAM, VPC, CloudWatch, and CodePipeline is genuinely deep.

The pricing is brutal. Their gpu1.large GPU runner is $0.30/min — that's $18/hr for a single GPU. The CPU sizes are $0.005–0.080/min, with no spot tier at any size.

machine.dev wins by 17× on GPU on-demand, 66× on GPU spot (CodeBuild has no GPU spot tier), 15.6× on small CPU spot, and 41% on small CPU on-demand. This is the most lopsided comparison in the entire market.

02 Cpu pricing
[REF_002] // LINUX_X64

CPU: Mostly A Sweep.

Per-minute USD. CodeBuild on-demand prices in us-east-1. CodeBuild has no spot tier at any size. Our biggest gaps are at small sizes; at very large sizes the on-demand math closes in.

[NOTE_SMALL] CodeBuild's small sizes (2–8 vCPU) carry a steep AWS managed-service markup. machine.dev runs a tighter 2× markup on EC2, which is why the on-demand gap is 41% at small sizes.

[NOTE_BIG] At 32–72 vCPU CodeBuild gets closer to raw EC2 rates and the on-demand gap closes. Spot still wins decisively (7–15×) because CodeBuild has no spot tier at any size.

02b Gpu pricing
[REF_003] // NVIDIA

$18/hr vs $0.27/hr.

CodeBuild's gpu1.large is $0.30/min on-demand — the highest published GPU rate in the entire market. machine.dev's T4 4-vCPU spot is $0.00451/min ($0.27/hr).

02c Features
[REF_004]

Beyond price.

SPEC CODEBUILD MACHINE_DEV
GPU on-demand pricing $0.30/min ($18/hr) — highest in market $0.01753/min on T4 4vCPU OD
GPU spot tier No (reserved capacity only) Yes (T4G ARM from $0.00277/min)
CPU spot tier No Yes (14-19× cheaper than OD)
CUDA pre-installed No (BYO image) Yes (12.1 + cuDNN 9.2.1)
Hosting Runs in your AWS account Fully managed
AWS account required Yes No
Setup CodeBuild project + IAM + buildspec GitHub App + runs-on change
GPU AMI maintenance Customer responsibility Managed
Lambda compute (small jobs) Yes (~$0.00004/sec/GB) No (use 2-vCPU spot at $0.00032/min)
Free tier 100 build min/mo on small $10 free compute, no card
03 What you keep
[REF_003] // SCOPE_OF_CHANGE

The only line you touch.

Your workflow files, steps, secrets, caches, and Actions syntax stay exactly as they are. The only change is the runs-on label.

BEFORE.YAML DEFAULT
jobs:
train:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- run: python train.py
AFTER.YAML EXECUTE
jobs:
train:
runs-on: [machine, gpu=T4]
steps:
- uses: actions/checkout@v4
- run: python train.py
04 Where codebuild wins
[REF_004]

Three strengths.

[REF_010] AWS // NATIVE

First-party AWS integration

CodeBuild runs in your AWS account with native IAM, VPC, CloudWatch, and CodePipeline integration. For shops already deep in AWS, the integration may justify the markup. Roles, policies, and audit trails all flow through your existing AWS controls.

[REF_011] LAMBDA

Lambda compute for very small jobs

CodeBuild Lambda compute mode can be very cheap for short jobs (~$0.00004/sec/GB). For sub-second tasks, the Lambda math can beat per-minute pricing. machine.dev doesn't compete in this mode.

[REF_012] RESERVED // GPU

Reserved capacity for big GPU

CodeBuild's reserved capacity supports the AWS G4/G5/P3/P4/A100 instance families — including A100 and H100 if your account has access. machine.dev's public catalog tops out at RTX 6000 (96 GB VRAM).

05 Where machine dev wins
[REF_005]

Five strengths.

[REF_020] PRICE // GPU_OD

17× cheaper on GPU on-demand

CodeBuild's gpu1.large is $0.30/min — that's $18/hr for a single GPU. machine.dev's T4 4-vCPU on-demand is $0.01753/min, ~17× cheaper. The CodeBuild GPU pricing is the most expensive published rate in the entire competitive landscape.

[REF_021] PRICE // GPU_SPOT

66× cheaper on GPU spot

CodeBuild has no GPU spot tier. machine.dev's T4 4-vCPU spot is $0.00451/min — 66× cheaper than CodeBuild's gpu1.large on-demand. T4G ARM 8-vCPU spot is $0.00277/min, 108× cheaper.

[REF_022] PRICE // CPU

15.6× cheaper on CPU spot, 41% cheaper on-demand at small sizes

CodeBuild's general1.small (2 vCPU) is $0.005/min. machine.dev's 2-vCPU spot is $0.00032/min and on-demand is $0.00297/min. The spot delta is 15.6× and the on-demand delta is 41% — both at small sizes where most CI jobs live.

[REF_023] CUDA // PRE_INSTALLED

CUDA pre-installed

CodeBuild GPU runners boot a generic AMI; you build your own image with NVIDIA drivers, CUDA, cuDNN, and your ML framework. machine.dev ships drivers 555.58, CUDA 12.1, and cuDNN 9.2.1 baked in. torch.cuda.is_available() returns True on the first run.

[REF_024] INFRA // ZERO

No AWS account required

CodeBuild requires an AWS account, an IAM role, a CodeBuild project, and a buildspec. machine.dev requires installing a GitHub App and changing a runs-on label. No AWS, no IAM, no buildspec.

06 Live pricing
[REF_PRICING] // LIVE_DATA

Live per-minute rates

See what you'd actually pay. Compare with GitHub's $0.006/min for 2-core, $0.012/min for 4-core, and $0.052/min for GPU.

07 Initialize
READY
> SYSTEM_READY // AWAITING_INPUT

Get started free.

$10 free compute on signup. Enough to run real workloads and compare cost before committing.