Billing
You pay for the minutes you use.
Credit is prepaid and metered by the minute against a per-card hourly rate. You see an estimate for each job from a probe of your own scene, you are charged the minutes you actually use, and there is no subscription.
01
Before you commit
Measuring your scene costs you nothing
We rent a GPU to probe your actual scene. That rental is ours, not yours — you are not billed for it whether or not you go ahead.
02
The rate and the estimate
A fixed rate, an honest estimate, a maximum
Before you approve, you see the job's hourly rate, an estimate of what the whole job will cost, and the most it can cost. The rate is fixed for your job, so a rate change after you approve cannot move it. The estimate is our forecast from the preview, not a promise — you pay for the minutes the job actually uses, up to the maximum.
03
The reservation
Credit is held, not spent
We move a little more than the estimate from available to held before renting anything. Holding is not charging: nothing has left your balance yet.
04
While it renders
Charged per minute, with a hard stop
The meter ticks at your job's fixed rate. If a render would run past the maximum you approved it is stopped and parked for a human, rather than billed onwards.
05
When it finishes
You are charged the actual, and the rest returns
The difference between what was held and what you used goes straight back to available credit. Unused balance stays yours for the next job.
Funding your balance
PayPal or crypto, both crediting one balance. A crypto deposit carries a 11.11% bonus because it costs us less to accept.
That is the only place the two differ. A render costs the same whichever way you funded it, because by the time you submit a job we have no idea which deposit is paying for it.
Which card, and what it costs
Among the cards that fit your scene's VRAM with headroom, we offer your job on the one that costs you least per frame — rate divided by speed, the last column below — comparing only the cards whose speed we have measured ourselves. Your job's rate is then that card's published rate, read from the same catalog the table is. You see it before you approve anything, and once fixed for your job it does not move with the rate card.
Each larger card costs more per unit of rendering, and only because it holds bigger scenes — so your scene is always offered on the smallest card that holds it, which is also the cheapest per frame.
| Card | VRAM | Rate | With VAT | Speed | Per RTX 4090 hour | Speed is |
|---|---|---|---|---|---|---|
| RTX 4090 | 24 GB | $3.30/hr | $3.99/hr incl. 21% IVA | 1.00× | $3.30 | reference card |
| RTX 5090 | 32 GB | $4.60/hr | $5.57/hr incl. 21% IVA | 1.30× | $3.54 | measured on our render path, 2026-08-19 |
| RTX PRO 5000 Blackwell | 48 GB | $5.40/hr | $6.53/hr incl. 21% IVA | 1.00× | $5.40 | measured on our render path, 2026-10-01 |
| RTX PRO 6000 Blackwell | 96 GB | $11.50/hr | $13.92/hr incl. 21% IVA | 1.40× | $8.21 | measured on our render path, 2026-09-28 |
Rates in USD, excluding VAT, per hour of the rented machine from the moment it reports ready until your job finishes, metered by the minute; the “with VAT” column adds the standard rate of the country picked above. “Per hour of RTX 4090 work” is the rate divided by the speed: what the same rendering costs on each card. These are the only cards a job runs on today: every preview runs on an RTX 4090, a scene that needs more memory is previewed again on a larger card, and the job renders where its preview ran — so every speed here is one we measured.
Among the cards whose speed we have measured, the RTX 4090 is the cheapest per unit of rendering. The RTX 5090 renders 1.373× as fast on our corpus but rents at $4.60/hr against $3.30/hr, so an hour of RTX 4090 work costs $3.54 on it — about 7% more per frame at the catalog's rounded speed of 1.30×. That is a fact about price per unit of rendering on our fleet at these rates, and nothing else about any card.
Every job renders on the card its preview ran on. The preview runs on an RTX 4090, and a scene whose preview uses up to 20 GB of GPU memory renders there. One that needs more is previewed again, still free, on an RTX 5090, and renders there at the RTX 5090's rate. One that needs more than about 26 GB is previewed again, still free, on an RTX PRO 5000 Blackwell, and renders there at the RTX PRO 5000 Blackwell's rate. One that needs more than about 40 GB is previewed again, still free, on an RTX PRO 6000 Blackwell, and renders there at the RTX PRO 6000 Blackwell's rate. Scenes whose preview needs more than about 80 GB of GPU memory are not supported yet. Finding that out costs you nothing: the preview is free, and nothing is charged.
Every card in the table has a speed we measured on our own render path, against the RTX 4090 that defines the scale. A rate is what you pay per hour of a card; a speed is how much rendering an hour of it does. A card whose speed we have not measured is never where a job runs, so it is not listed.
Speed relative to an RTX 4090, measured on our own render path rather than taken from a published benchmark. How we measure.
How rates are set
One published rate per card, the same for everyone: no priority tiers, no volume pricing, and no billing per benchmark point. How you fund your balance never changes what a render costs.
We review rates at most once a month — at least 28 days between changes to any card — against what the hardware costs us and what comparable services charge. A rate change never touches a job already quoted: it keeps the rate you approved.
Every change to a published rate will be listed here, with its date.
If something goes wrong
A machine disappears
Frames are written as they finish, so a reclaimed machine costs at most the frame in flight. The job resumes on a new one from where it stopped, and you are not charged twice for the same frame.
A render cannot finish
You are charged for what was metered and no more, and any frames it did produce are yours to download. A job that stops gives back both the hardware and the unused credit.