Reputation & Uptime

Every card on the Rent page is someone else's machine. Before you put a job on one, you want to know how it behaves when it's actually rented — not how long it sits idle looking available. That's what reputation measures.

What the score is built from

A GPU's reputation comes only from finished, paid rentals — real sessions that ran and ended. Idle uptime doesn't count: a machine that sits online all day tells you nothing about whether it will hold up under your workload. Each completed rental records how quickly the machine started and whether it stayed healthy for the whole session, and those roll up into a rolling score. Recent rentals count for more than old ones, so a score reflects how a machine behaves now.

Two scores: the machine and the provider

You'll see up to two reputation signals on a card:

• Node score — this specific machine. A star with a number (for example ★ 98) when it has enough history to be proven.
• Provider score — the operator behind it, across every machine they run (for example Provider 94). A provider builds this up over their whole fleet, so it tells you about the person you're renting from, not just one card.

Both matter. A brand-new machine from a long-proven provider is a very different bet from a brand-new machine from someone with no track record.

Only real host problems count against a machine

This is the part that keeps the score honest: a rental is judged on the resource, not on you. If your own job crashes — an out-of-memory error, a bad script, a process that exits with an error while the machine itself stayed online and healthy — that does not lower the machine's score. It isn't the provider's fault, so it isn't counted against them.

What does lower a score is a host failure: the machine going offline while your rental was still running, or never coming up at all. That's the provider's responsibility, and it's exactly what reputation is there to surface. Anything genuinely ambiguous is set aside rather than blamed on either side.

The three tiers

• New — not enough completed rentals yet to earn a score. A new machine isn't a bad machine; it's an unproven one. You'll still see its verified specs and hardware — reputation is earned, never faked with a made-up rating.
• Proven — enough history to trust. Shown with its score and its measured uptime while rented.
• Flagged — repeated host-caused failures. Surfaced with a warning so you can steer around it.

Where you'll see it

Reputation shows up on each card and on the GPU's own page on the Rent page — the node score, the provider score, and, for a proven machine, its uptime while rented and how quickly it typically starts. Use it the way you'd use a rating anywhere else: as one input alongside the GPU model, VRAM, price and location.

If you're a provider

Your machines earn reputation the same way — by serving clean rentals. Two things worth knowing:

• Your provider score follows your account, across every node you run. It isn't tied to a single machine, so it reflects your track record as an operator.
• You're only judged on what you control. A renter's own crashed job never counts against you. What counts is keeping a machine up and healthy for the duration of a rental — that's the commitment renting makes on your behalf, and reputation is how you get credit for keeping it.

See Add a Node to get started and Node Pricing to set your rate.

Questions? Join the Discord — we're happy to walk through how a specific machine or provider is scored.