A Managed Core Hour is one worker CPU core, managed by Orion, for one hour. It's the only unit Orion is priced on. We only charge for what Juno is managing, so the count covers the physical CPU cores on the worker nodes that run your workloads. Management nodes, threads, GPUs, servers and users are never counted.
What counts as a core?
Orion counts physical cores on worker nodes and nothing else. The full definition is in our methodology glossary, and the rules are short.
Worker nodes count. Management nodes, the machines that run Orion itself, are free.
Physical cores count, never threads. A 48-core node with 96 threads counts as 48.
There are no server counts and no user counts.
GPUs are never licensed. Every GPU is included at no extra charge.
How do you count your own cores?
Start with the worker nodes you want Orion to manage, and leave the management nodes off the list. On a Linux node, lscpu shows Socket(s) and Core(s) per socket. Multiply the two and you have the physical core count for that machine. Ignore Thread(s) per core.
If you don't have shell access to every machine, the make and model is enough, and we can look up the core count from there. A node counts once, however many people or teams share it, so you need one row per machine.
What does that look like on a real cluster?
Take a small cluster with three management nodes running Orion and four worker nodes running the workloads. Two of the workers are 48-core servers with 96 threads each, and each one drives four GPUs. The other two are 32-core servers with no GPUs.
The count is 48 plus 48 plus 32 plus 32, which is 160 managed cores. The management nodes, the extra threads and the eight GPUs add nothing to it. That's it.
Small machines work the same way. A DGX Spark counts its 20 Grace cores, whatever you run on its GPU. We wrote about running a Spark for a whole team in this post.
Why cores and not seats or GPUs?
Seats change every semester and every hire. Somebody has to keep counting them, and somebody else has to argue about whether the intern who logged in twice is a seat.
GPUs are the hardware you most want people to share. Orion exists to get more people onto the GPUs you already own, so charging per GPU would charge you more for the thing we're trying to help you do. A node with 32 CPU cores driving eight GPUs counts as 32.
Cores sit on the hardware list. They don't move unless you buy hardware, and anyone with that list can check the count.
Do you count hours one by one?
Nobody counts hours on a subscription, even though the hour is the unit. Your quote is one annual number for the cores Orion manages over the term. It doesn't go up in a busy month or down in a quiet one.
What happens if the cluster grows past the license?
Orion never kills a workload. If you add nodes mid-year because a grant landed or a team joined, everything keeps running, and we true up at renewal.
That's on purpose. Nobody wants to find out their license ran short because a user's job stopped halfway through.
How does the license work on an air-gapped cluster?
Orion is built to run air-gapped, and the license works the same way connected or offline. You buy it up front, and it arrives as a signed license file that Orion checks locally. Orion never calls home, and no logs leave the site to count cores.
At renewal you tell us your peak core count and we true up from there. A disconnected site pays the same price as a connected one.
How do you buy Managed Core Hours?
There are two ways to buy. A subscription runs a year at a time. A multi-year term commits you for longer and brings the rate down. Both include every feature, updates and Standard support.
We don't publish a rate. The number depends on your core count and your term, and I'd rather send you the total for your cluster than a rate you have to multiply out yourself.
Can you try Orion without buying anything?
You can. Community Edition is free to install and runs two workloads at a time, with community support and no PO. When you want to test properly on your own cluster, a paid Proof of Value covers that, and part of the fee comes off year one if you sign after the readout.
Count the physical cores on your worker nodes and you already have most of a quote. The pricing page walks through a sample cluster.
WRITTEN BY
Tony Como
Tony runs go-to-market at Juno, from first outreach to production deployment. He works across sales, partnerships, and operations to help technical teams move faster without adding infrastructure overhead.