Fast GitHub Actions runners on AMD Zen 5
Change one line of your workflow. Every job gets its own fresh VM on a desktop-class AMD Zen 5 CPU, and our own CI jobs finish in 20 to 111 seconds.
Install the GitHub App Read the quickstart
jobs:
test:
- runs-on: ubuntu-latest
+ runs-on: coreloadBeta: install the App, and we turn your organization on by hand and send you its label.
- 20 to 111 seconds. Each job of Coreload's own CI, on Coreload runners, as of 2026-11-03. How we measured
- AMD Zen 5 desktop CPUs. High clocks and strong single-thread speed, which is what most CI steps wait on.
- A fresh VM for every job. It never serves a second job, and its disk and memory are deleted when the job ends.
Measured speed
What we have measured so far is our own CI: each of its jobs takes 20 to 111 seconds on Coreload runners.
Method. The wall-clock time of each job, from start to finish, in Coreload's own repository, as of 2026-11-03. The per-size comparison against Blacksmith, with its raw results and a benchmark you can run yourself, is on the roadmap below.
How it works
- Install the GitHub App. Pick the repositories it may serve. During the beta, we turn your organization on by hand and send you its label; jobs wait until then.
- Change one line. Set
runs-on:to the label we send you on the jobs you want faster. Jobs you don't change keep running where they ran before. - Each job gets its own VM. A fresh VM starts for the job, runs it, and is deleted when it ends.
Sizes, one label each
runs-on: | Machine | Runs on | Price |
|---|---|---|---|
coreload | 2 vCPU · 7 GiB | Rise-S, Rise-M, Rise-L | $0.00114 / min |
coreload-4vcpu | 4 vCPU · 12 GiB | Rise-S, Rise-M, Rise-L | $0.00228 / min |
coreload-8vcpu | 8 vCPU · 32 GiB | Rise-S, Rise-M, Rise-L | $0.00418 / min |
Beta pricing: published before we bill anything
New organizations start at 2 vCPU.
Security and isolation
- One VM per job. Each job runs in a fresh VM that never serves a second job or a second customer. Its disk and memory are deleted with it when the job ends.
- We don't store your secrets. A secret your workflow passes to a job is visible inside that job's VM while it runs, so pass short-lived, least-privilege credentials.
- The App asks for little. It registers single-use runners in its own runner group, and has no access to your repository contents, issues, pull requests or secret values.
- Where jobs run. On bare-metal machines we rent from OVHcloud. Our control plane runs on AWS.
Pricing
Coreload is in beta. Pricing is published before we bill anything.
Questions
- How do I switch from GitHub-hosted runners? Install the GitHub App, then change
runs-onon a job to the label we send you. Jobs you don't change keep running where they ran before. - How do I switch from Blacksmith? Change
runs-onfrom your Blacksmith label to the label we send you, matching sizes by vCPU; new organizations start at 2 vCPU. Install the GitHub App first, then follow the migration guide. - What is a vCPU on Coreload? One CPU of the job's VM, on an AMD Zen 5 desktop machine. We will state whether it is a hardware thread or a core when the per-size benchmark is published.
- Which sizes are there? The sizes in the size table. New organizations start at 2 vCPU.
- Where do my jobs run? On bare-metal machines we rent from OVHcloud, each job in a fresh VM that is deleted when the job ends.
- Personal accounts? Not yet: organizations only, private repositories only.
- What does it cost? Pricing is published before we bill anything.
On the roadmap coming soon
Coming soon: not available yet.
- One label,
coreload, for every organization, with no manual turn-on. - 4 and 8 vCPU for every organization, then 16 and 32 vCPU, in the same shapes as Blacksmith.
- An Ubuntu 24.04 image like GitHub's, with Docker.
- A Coreload cache next to the machines, with nothing to set up: dedicated machines first, then the shared pool.
- Per-second billing from a prepaid balance.
- A public status page.
- A per-size speed comparison against Blacksmith, with a benchmark you can run.