Dedicated Instant Custom GPU Cloud VPS Unmetered Domains Pricing Data Centers Migrate
Marketplace Configure server
GPU server marketplace · for people who own the hardware

List the GPU you own. Earn every month. Stay in control.

This is the selling side of the trade. You already have the card and the machine; the marketplace connects them to AI teams, studios and engineers who need GPU access, and handles verification, listing and demand routing while you set the price and the availability. Looking to rent a GPU server from us instead? That is GPU dedicated servers, and it is a different page.

Submit → verify → publish Simple monthly billing Monitoring & isolation
List your GPU See what other hosts are asking Data centres: list a fleet
Performance benchmarking before every listing You control pricing and availability windows Your tenants come from AI, inference, rendering and media

How the marketplace works

A straightforward pipeline from submission to revenue: apply, verify, publish, earn. You retain control over your hardware at every stage.

Host onboarding

  1. Submit your GPU specifications along with a management access method (SSH or IPMI).
  2. Verification: we run driver health checks, VRAM validation, ECC status, PCIe lane checks, and thermal stress tests.
  3. Performance: sample inference and training workloads plus network and storage throughput benchmarks.
  4. Publish: your listing goes live with a Verified badge and the monthly asking price you configure.
  5. Earn: renters are provisioned on demand and pay your monthly asking price.

As community hosts submit machines, approved configurations will appear here after testing and verification.

Verification checklist

  • GPU

    Model identification, VRAM capacity, clock speeds, thermals, persistence mode, and load stability.

  • CPU & RAM

    Baseline compute and memory throughput to rule out bottlenecks.

  • Storage

    Sequential and random read/write speeds plus IOPS for dataset-intensive workloads.

  • Network

    Latency, throughput, packet-loss screening, and routing verification.

List your GPU from your dashboard

Everything happens in your Primcast dashboard — install the lightweight agent, register your machine with a one-time code, and publish your listing once it passes verification. Set a monthly asking price and stay in full control.

Three steps, all in the dashboard

  1. Install the agent on your GPU host using the one-line command shown on your dashboard.
  2. Register & generate a code to securely link the machine to your account.
  3. Publish: set your monthly asking price and go live after verification.
Open dashboard → List a GPU

Why list from the dashboard

  • Full control

    Update pricing, availability, and specs whenever you want.

  • Live status

    Track verification, active rentals, and payouts in one place.

  • Secure linking

    Agent plus a one-time code — no need to hand over standing SSH credentials.

  • Instant edits

    Pause, relist, or reprice without contacting support.

Data centres: list a whole fleet

Racking tens or hundreds of GPU machines? A data centre account lists them all with one command, puts your company's name on every machine, and gives renters your service level instead of the community terms.

What a data centre account adds

  • One command for every machine

    A fleet code registers up to 1,000 machines and asks nothing, so the same line runs across your racks from Ansible, cloud-init or a loop over SSH. It can publish each machine at your price and only accept your own IP ranges.

  • The Data centre partner label

    Once we verify your company, renters see your name and data centre on each machine instead of "Community host".

  • Your service level, in writing

    Your hardware replacement window and a network credit of one day for every hour a machine is down, shown on the order page and on the renter's machine page.

  • Business billing

    A monthly statement per machine between our US company and yours, in US dollars, with rented hours, our fee and your net.

How it starts

  1. Create the account: sign in or sign up, then tell us your company, where the machines are racked and roughly how many you will list. Your first 10 machines can go on straight away.
  2. We review it and verify your company, usually within one business day.
  3. List the rest: fleet codes of up to 1,000 machines, with the label and your service level on every one.
Create a data centre account

Network and security

What your connection needs, and what a renter can and cannot reach on your machine.

Your network

  • Outbound only

    The agent makes one outbound SSH connection to our relay (TCP port 2222), and renters reach your machine through that tunnel. Nothing is opened on your router, so you do not need a public IP, and a connection behind CGNAT works.

  • No minimum upload speed

    We do not set a minimum upload speed. The agent measures download, upload and latency to the speed test server in your location and shows them on your listing, and verification checks the network path. Renters see those figures, so a faster line makes the listing more attractive.

What a renter can reach

  • A machine of their own

    Each rental runs in its own virtual machine with the GPU passed through to it. An NVIDIA GB10 (DGX Spark, ASUS Ascent GX10, ZGX nano) is rented as a hardened container instead, because its GPU cannot be passed through to a virtual machine yet: the renter logs in as an ordinary user without sudo, root inside the container is an unprivileged user on your machine, and almost every privilege is removed.

  • Your files stay yours

    The renter cannot see your files. What they write goes to a separate encrypted volume whose key exists only in memory and is destroyed when the rental ends.

  • Your network is fenced off

    Before a rental starts, the agent loads a firewall table and checks it. That table blocks the renter from your home network, from every network configured on your machine, from private address ranges and from the machine itself, and nothing on your network can connect in to the renter.

  • Checked before it is rented again

    If the cleanup or the GPU check at the end of a rental cannot be verified, the machine is taken off the market rather than rented again.

The shelf your listing joins

Before you set an asking price, it is worth seeing what the machine next to yours is asking. These are the listings already live, and the Primcast machines sitting on the same shelf.

What Primcast itself lists

Our own machines, on the same shelf and priced the same way — the figures a host is pricing against.
Instant servers is where these are ordered, and GPU dedicated servers prices every card we fit.

Listed by their owners, right now

Every machine here belongs to somebody who listed it, at the monthly figure that person set. Filter it the way a prospective tenant will, and you can see where your own listing would land.

Loading…

Earnings estimator

Project your monthly payout based on your asking price.

Estimated monthly payout $

What drives higher earnings

  • Competitive pricing

    A well-priced listing rents out faster and stays rented longer.

  • Consistent availability

    Always-on listings attract more long-running rental contracts.

  • Fast I/O

    NVMe storage and stable networking improve renter retention.

  • Verified badge

    Verified machines receive priority placement and higher conversion rates.

What your machine needs

Install nothing but the agent. There is no Docker, CUDA or ML framework to set up on your side: once you link the machine, the agent installs its own rental runtime, builds the rental image and runs a test rental by itself. Renters get a clean Ubuntu machine with the GPU and its driver, and install the CUDA toolkit, PyTorch or TensorFlow they want.

64-bit LinuxUbuntu, Debian, DGX OS or Armbian, on Intel/AMD or ARM. Windows and macOS hosts are coming later. OS
KVM virtualisationOn in the BIOS (Intel VT-x or AMD-V). Stock Ubuntu and Debian kernels have everything else it needs. Host
IOMMU onVT-d or AMD-Vi (the SMMU on ARM), so the GPU can be passed to the rental whole. Not needed without a GPU. Firmware
Any GPU, or noneNVIDIA, AMD or Intel, with no driver needed on the host. A machine without a GPU rents out its CPUs. GPU
4 GB memory, 40 GB diskAt least, free on internal storage (never an SD card). A rental gets everything the machine does not keep for itself. Resources
Outbound internetOne outbound connection to our relay on TCP port 2222. No public IP or open port; CGNAT works. Network
No desktop on the GPUA workstation runs headless while listed, and the agent can switch it. A DGX Spark keeps its desktop until it is rented. Host
Installed for youThe QEMU/KVM runtime, an Ubuntu 26.04 rental image with the NVIDIA driver to match, and a test rental. About 25 to 45 minutes after linking. Automatic
DGX Spark and GB10Rented as a hardened container on the machine's own NVIDIA driver, with podman and the NVIDIA Container Toolkit set up by the agent. Automatic

Frequently asked questions

What a host needs to know before listing a machine, including how the tenants get into it.

How are hosts paid?

You earn for every hour a renter has your machine and it is online: your monthly asking price divided by 720, less the 12% platform fee. Earnings are paid monthly in USD by PayPal, within the first ten days of the following month, once your balance reaches $50; a smaller balance carries over. Identity verification is required before the first payout. The Earnings tab in your dashboard shows what has accrued, what has been paid and what is still owed.

Which fees does a host pay?

The 12% platform fee is the only amount Primcast deducts from your earnings. The renter pays a separate 5% platform fee on top of your asking price, and it never comes out of yours. We pay what PayPal charges us to send your payout. Anything PayPal charges you to receive it, or to convert it into your currency, is PayPal's fee, set and charged by PayPal, not by us.

Where do I enter my PayPal account?

Under List a GPU in your dashboard, the Payout details card takes your PayPal email address, the account holder's name, your country, and whether you are a private individual or a company. Every payout goes to that address until you change it there.

Can I host from any country?

Hosts are not limited to a list of countries. Three things have to hold: your machine can keep an outbound connection to our relay on TCP port 2222 (no public IP or open port is needed, and CGNAT works), you pass identity verification before your first payout, and your PayPal account can receive USD in your country. Where PayPal cannot pay you, neither can we.

What happens if verification fails?

You receive a full breakdown of every failed check — thermal limits, driver errors, storage throughput, or network issues. Fix the flagged items and resubmit; there is no limit on retries.

Which software environments are compatible?

Anything that runs on Linux with a GPU. Each rental is a private Ubuntu 26.04 machine with the GPU passed through and its driver installed; renters add the CUDA toolkit, ROCm, PyTorch, TensorFlow or Docker they want. On a DGX Spark the rental is a hardened container that uses the machine's own NVIDIA driver. Hosts install nothing but the agent: before a machine is listed, it builds the rental image and runs a test rental that proves the GPU reaches it.

Is datacenter-grade hardware required?

It is not mandatory, however stable uptime, adequate cooling, and reliable connectivity are non-negotiable. Hosts running from professional facilities generally see higher occupancy and fewer disruptions.

How are renters given access?

Your machine runs a private virtual machine for the renter, with the GPU passed through and their SSH key inside. They reach it over SSH through our relay: no IPMI, no login to your system.

How long before my listing goes live?

As soon as you price it. A newly registered machine stays private until you enter its monthly asking price in your dashboard; from then on renters see it whenever it is online and passing its hosting checks. Nothing waits on our approval, and a failing check shows up in your dashboard with what to fix.

Have a GPU sitting idle?

List it, set the monthly figure you want for it, and keep the machine. Verification is ours; the price and the availability stay yours.

List your GPU

Racking many machines in a data centre? Create a data centre account and list the whole fleet with one command.

Came here to rent GPU compute rather than let it? That is the other direction of this trade: GPU dedicated servers prices every card we fit, and instant servers is what is racked right now.

Own an NVIDIA DGX Spark, an ASUS Ascent GX10 or a ZGX nano? Those have their own page, because an arm64 GB10 box lists on different terms from a card in a chassis: rent out a GB10 machine you own. To rent one from us instead, that is the DGX Spark dedicated server.