OctaSpace calls a listed machine a node, and the word is doing real work: a renter takes the whole node rather than a slice of it. That changes the setup in ways that catch hosts arriving from Vast.ai, starting with a filesystem requirement most people have never met before. Here is the path from bare hardware to a verified, earning node.
Check these details against the source
Hardware: what each node type needs
OctaSpace runs three kinds of node, and the requirements differ by quite a lot. Their overview mixes preferred figures with stated minimums, so the column below says which each one is:
| Node type | What it serves | Stated requirement |
|---|---|---|
| Compute | General GPU workloads | Preferred: 8 CPU cores, 32 GB RAM, 500 GB free disk, NVIDIA RTX 30-series or better |
| Rendering | OctaRender jobs | Minimum RAM: twice VRAM multiplied by GPU count. Optimal CPU: 8 cores per GPU |
| VPN | Network services | Minimum: 1 CPU core, 1 GB RAM, 10 GB disk, and a static IPv4 address |
The rendering rule is the one worth doing the arithmetic on before you buy anything, and note that it replaces the compute RAM figure rather than adding to it. Four RTX 4090s at 24 GB each is 96 GB of VRAM, which by that formula wants 192 GB of system RAM. The RAM is a stated minimum. The 8-cores-per-GPU figure is not: OctaSpace calls it optimal and says lower core counts are still supported, they just render less efficiently. So 32 cores is what a four-card render node wants, not what it needs to qualify.
Operating system and drivers
OctaSpace's node software runs on Linux and supports both x86 and arm64. Their documentation names Ubuntu 24.04 Server as the preferred release, which lines up with the recommendation in our Ubuntu server setup checklist, so that guide applies here more or less as written.
Two driver details that are specific to OctaSpace:
- NVIDIA driver 550 or newer. Their stated minimum. Note this is a different number from the branch 570 that RTX 50-series cards need for the silicon itself, so on a 5090 node the card's requirement is the binding one.
- Matching minor versions. The driver and the
libnvidia-glorlibnvidia-gl-serverpackages must share the same minor version. A mismatch here is a classic silent failure: the machine looks installed and the node does not work.
The disk requirement is unusual: xfs on /var/lib/docker
xfs and mounted at /var/lib/docker. This is not how a default Ubuntu install arrives, and it is not something you can retrofit casually once containers are running. Install xfsprogs with apt install -y xfsprogs and sort the filesystem out before the node software, not after. Their installer will claim 80% of the free space on the root filesystem for Docker data unless you set export SKIP_FS_CONFIG=1 first. The variable name is uppercase and shell variables are case sensitive, so a lowercase copy of it does nothing and you get the automatic setup anyway.Install and register the node
- 1As root, run the installer:
curl https://install.octa.space | bash - 2Decide whether to configure service ports. OctaSpace routes traffic through its own proxy servers when you do not, so this step is optional unless you are providing a VPN service. Configuring them gives better network speed and more direct access, which is worth having. It also needs a static public IP address, so if you do not have one, take the proxy default and move on.
- 3If you are configuring them: forward ports 51800 to 51816, TCP and UDP both. That range is the default and it is narrow, which makes it easier to get right than a Vast.ai-style block of hundreds.
- 4Take the node token the installer gives you and register the machine at cube.octa.computer under Hosting, then Add Node.
- 5Set a Reward Address in your profile settings. This is not bookkeeping you can leave for later: OctaSpace's documentation states the node is not made available on the marketplace at all until one is set.
- 6Check the node dashboard on port 18888 to confirm the machine is reporting.
If Docker refuses to start after installation, OctaSpace's docs point at IPv6 needing to be enabled as a known cause. That one is worth remembering, because nothing about the symptom suggests the cause.
Verification: four checks, and they are re-run
An OctaSpace node is either verified or unverified, and the difference is not cosmetic. Their documentation states that unverified nodes cannot deliver services or take part in staking, so an unverified node earns nothing at all rather than simply earning less.
Verification checks four things:
- The system clock is synchronised.
- All the necessary network ports are open.
- The GPU driver is correctly installed.
- The hardware requirements are met.
If a node fails, you fix the underlying problem and restart the OSN service to trigger a re-check. The part worth internalising is that verified nodes are re-checked periodically rather than being verified once and left alone. A clock that drifts, a driver that a background upgrade replaced, or, if you configured service ports, a forward that a router firmware update quietly dropped can all take a working node out of service without any visible event. That is a strong argument for turning off unattended upgrades, covered in why auto updates kill GPU host uptime, and for testing your ports from outside the network, covered in port forwarding for GPU hosts.
Idle jobs: what to do with the hours nobody rents
OctaSpace lets a node run background work while it is not rented. You configure a Docker image with its entrypoint arguments, environment variables, and an optional startup script; up to three jobs can be configured per node, and they can be scheduled by day and time. The documented use is mining, and the platform will start and stop a HiveOS flight sheet automatically if you run one.
Treat this as a different question from pricing rather than a substitute for it. Idle jobs monetise hours your rate failed to sell, which is worth something, but a node that idles enough for them to matter is usually a node that is priced wrong. The revenue arithmetic is in why utilization beats listing price.
One thing their docs do not answer
Next: the pricing model is the part that is genuinely different
Setup on OctaSpace is unusual mostly in its details. Pricing is unusual in its shape, because selling a whole node is a different business from selling GPU slots, and the limit it puts on your per-GPU rate surprises multi-GPU hosts in particular. That is covered in whole-node pricing on OctaSpace, which is the guide to read before you set your first price.