Render farm pricing uses four common billing units: GHz-hours for CPU work, OctaneBench-hours (OB-hours) for GPU work, node-hours for whole render servers, and machine-hours for GPU servers you rent and run yourself. The listed rate only becomes comparable once you convert it to cost per frame for your own scene. Multiply the frame time by the unit the farm counts, apply the rate for your priority, then add VAT, minimums, idle time and any storage or software charges the farm lists.

• Same job, different units: iRender’s measured 300-frame Octane job on 4x RTX 5090 (5.23 card-hours) costs about $59 on iRender and between about $11 and $108 at other farms’ listed rates, before VAT.
• OB-hours convert to card-hours: OTOY lists one RTX 5090 at 1,730 in OctaneBench 2025.2, so $0.003 per OB-hour works out to about $5.19 per card-hour.
• GHz-hour rates are not like for like: farms may count the GHz in a node differently and rarely say how, so a lower GHz-hour rate does not prove a lower bill.
• Rented machines bill while switched on: setup, scene checks and idle time count on machine-hour services, which matters most for short jobs.
• Measured jobs on iRender: iRender’s own published Octane and Redshift benchmarks cost about $0.06 to $0.43 per frame at its hourly rates, roughly 1.7 to 2.1 times the same card-hours at a $0.003 OB-hour rate.
• What to do next: render three sample frames locally, note the frame time and the hardware, then run the numbers in a cost calculator for two or three farms.

A client wants a 300-frame walkthrough by Friday, and you have four render farm pricing pages open. One quotes GHz-hours, one OB-hours, one node-hours and one an hourly rate for a whole GPU server. None of those numbers answers the question the client asked: what will this cost?

This guide explains how render farm pricing works for each unit, then converts all of them to cost per frame using rates listed on each farm’s page on 9 November 2026. The worked examples use one measured job published by iRender plus stated assumptions, so you can repeat the method with your own scene.

1. Why is render farm pricing so hard to compare?

Render farm pricing is hard to compare because farms bill in different units. CPU farms usually count GHz-hours, GPU farms count OctaneBench-hours, some count whole nodes, and rental services bill the hours a machine is switched on. A rate only means something once you turn it into cost per frame.

1.1. The four billing units at a glance

The table below lists the unit each farm uses, sorted alphabetically. Several farms use two units at once, GHz-hours for CPU and OB-hours for GPU. Others sell a whole server by the hour. Because of that, a lower number on one page can still mean a higher bill.

1.2. What the listed rate leaves out

A pricing page also states, or leaves out, several things that move the final total. These include tax, the priority level, whether waiting time is billed and whether idle time counts. Section 6 covers them. For now, keep one rule in mind: compare totals for a known job, not rates.

2. How do GHz-hours work on CPU render farms?

A GHz-hour is one CPU core running at 1 GHz for one hour. The farm multiplies the cores in the node by their clock speed and by the render time, then charges a rate per GHz-hour. It suits CPU engines such as Corona and V-Ray CPU, where many cores work on one frame.

2.1. The formula, with a worked example

Super Renders Farm’s pricing documentation gives the formula as cores x clock speed x render time. Its own example is a frame that takes 15 minutes on a 64-core machine at 3.5 GHz. That frame uses 64 x 3.5 x 0.25 = 56 GHz-hours. At its Standard rate, listed from $0.004, the frame costs $0.224.

The same 56 GHz-hours would cost $0.79 at RebusFarm’s listed 1.41 cents (read as US cents), or €0.62 at Ranch Computing’s low-priority €0.011. However, those three bills are not for the same work. That is the catch with GHz-hour render farm pricing.

2.2. Why a cheaper GHz-hour rate does not prove a cheaper bill

The GHz figure is a clock speed, not a measure of work done. A modern core at 3.5 GHz finishes more work per hour than an older core at the same clock. Farms also count differently. The pages rarely explain how they arrive at the GHz figure for a node, for example whether they use base or boost clock.

GarageFarm, for example, lists CPU rates from $0.024 to $0.072 per GHz-hour. That is six times Super Renders Farm’s $0.004 at the low end. The pages do not publish enough detail to say whether the two farms count the same GHz in the same way. So a GHz-hour rate is only comparable inside one farm, between its own priority levels.

If you render Corona or V-Ray CPU scenes, the Corona render farm guide and the V-Ray render farm guide list which farms support each host app.

3. How do OB-hours work on GPU render farms?

An OB-hour is GPU time weighted by the card’s OctaneBench score. A card that scores 1,730 uses 1,730 OB-hours in one hour of rendering. The farm multiplies those OB-hours by its rate, so faster cards cost more per hour but finish sooner.

3.1. What OctaneBench measures

OctaneBench is OTOY’s benchmark for OctaneRender. Its results page scores each GPU relative to a GTX 980, which sits at 100. On 9 November 2026 the page listed OctaneBench 2025.2 scores of 1,730 for one RTX 5090 and 1,307 for one RTX 4090. It also listed 1,641 for an RTX PRO 6000 Blackwell Server Edition.

3.2. Converting an OB-hour rate into a card-hour price

Multiply the OB-hour rate by the card’s score to get the cost of one card for one hour. At Super Renders Farm’s $0.003 per OB-hour, one RTX 5090 costs about $5.19 per hour. Super Renders Farm lists $5.17 per hour for one RTX 5090, close to that figure. Its own documentation counts the card as 1,731 OB, which shows that farms can use slightly different scores for the same GPU.

The same conversion gives about $6.92 per RTX 5090 hour at GarageFarm’s low priority and about $9.17 at RebusFarm. Ranch Computing’s low-priority €0.005 gives about €8.65. These figures assume every farm scores the card at 1,730. Super Renders Farm’s documentation and GarageFarm’s pricing page both name OctaneBench 2025.2.1; the other pages do not state a version. RebusFarm lists its rate as “0.53 cent/OBh” without saying whether the cent is in US dollars or RenderPoints, so these figures assume US dollars.

3.3. Where OB-hours mislead

OctaneBench measures Octane. Redshift, Cycles and V-Ray GPU scale differently from card to card. As a result, an OB-hour price is a fair way to bill hardware, but a rough way to predict your own frame time. For Redshift or Blender jobs, see the Redshift render farm guide and the Blender render farm guide.

4. What does node-hour pricing mean?

Node-hour pricing charges a flat rate for each render server per hour, whatever its benchmark score. The bill is nodes x hours x rate. To compare it with GHz-hour or OB-hour pricing, you need the node’s exact hardware, because that decides how many frames it finishes in an hour.

4.1. Two examples from pricing pages

Fox Renderfarm lists CPU node-hours from $1.224 (Ordinary) down to $0.734 (Diamond, for accounts that have recharged over $10,000). Its page states that each node includes 64GB of RAM by default and that larger RAM costs extra. It also notes that some rendering software carries additional charges, shown when you submit a task.

AnimaRender sells server-hours in AnimaPoints, where 1 AP costs $1. It lists $6.20 per hour for a Dual AMD EPYC 7642 server and $4.95 for a Threadripper 3970X server, billed per second. The page also states that the final cost of a render does not depend on server type, so ask which figure applies to your job.

4.2. What to check before trusting a node-hour rate

Ask for the CPU or GPU model, core count, RAM and the number of nodes your job can use at once. A low node-hour rate on older hardware can cost more per frame than a higher rate on newer hardware. Render one sample frame on the trial before you budget the rest.

5. How does per-machine billing work on a rented GPU server?

Per-machine billing charges an hourly rate for a whole GPU server that you start, use through remote desktop and shut down yourself. You pay while it runs, so scene checks and idle time count. In return, you control the software, the plugins and the render settings.

5.1. Two services that bill this way

Xesktop lists one RTX PRO 6000 Blackwell (96GB) at $1.99 per hour and two cards at $3.98, with per-second billing and no monthly minimum. It says the server is billed only while it runs, and its homepage lists volume discounts down to $0.99 per hour.

iRender’s pricing page lists RTX 5090 servers with 1, 2, 4 or 8 cards, from $10.8 per hour for one card to $62 for eight. Paying for 3 hours or more lowers those rates by about 10%, for example to $9.72 and $55.80. RTX 4090 servers start at $8.2 per hour. iRender’s pricing page lists its data center as Uptime Tier III with ISO 27001, ISO 20000 and ISO 9001 certifications, and the company provides an NDA document for projects that need one (checked November 2026).

5.2. How billing and saved machines work on iRender

iRender’s user documentation sets out the cycle. Billing on pay-as-you-go runs from the moment the machine starts running until you press Shutdown, and the documentation states that the boot step is not charged. Boot takes about 7 to 10 minutes for a machine under 50GB, 10 to 15 minutes between 50GB and 100GB, and more than 18 minutes above that.

The machine itself persists between sessions. According to the documentation, iRender saves the working environment after each session, including installed software, settings and files kept on the local C drive. On the next boot you continue in the same setup, so software is installed once, not every session. The documentation recommends keeping only software on the C drive, since a larger machine boots more slowly.

What does not survive is work that never reached a disk. Anything still unsaved in an open application is gone after shutdown, and iRender’s FAQ adds that shutdown purges temporary data caches. The documentation also warns that a forced Reset may cause data loss. It advises saving render output to the C drive, because output written to a network drive may not save if the connection drops. Results then go to the OUTPUT (Y:) drive before you shut down, so you can download them.

5.3. The limitation that matters for cost

The meter runs whenever the machine is on, whether or not it is rendering. Forgetting to shut down after an overnight batch costs real money, so the platform offers a Schedule Shutdown option. The video below walks through the start, render and shutdown cycle.

5.4. What the higher hourly rate buys

A rented server earns its cost when the job needs software an automated farm will not run, or when the work is interactive. Lumion, Enscape, D5 Render and Twinmotion are real-time renderers that run on a workstation GPU. Lumion’s knowledge base states that conventional render node functionality is not available in Lumion, so a full remote machine is the practical route for it. Check the other apps’ own cloud guidance before you choose. For a long Corona or V-Ray CPU batch, a GHz-hour farm that bills only render time may cost less. The table below sets out what each model gives you for the money.

6. How do you turn any unit into cost per frame?

Take one frame time from a test, convert it into the farm’s unit, apply the rate, add overheads, then divide by the frame count. The table below takes one measured GPU job, a 300-frame Octane sequence that iRender rendered on its own 4x RTX 5090 machine, and prices the same work at each farm. It shows how far render farm pricing can spread for the same work.

6.1. One job priced at listed rates

The table holds the frame time fixed and changes only the rate and the unit. In a real comparison, the frame time changes too, because each farm runs different hardware. Use it as a template rather than a ranking.

6.2. Why eight GPUs are not eight times faster

Multi-GPU rendering scales below linear, and how far below depends on the engine. In Puget Systems’ 2022 test with up to seven RTX 4090 cards, OctaneBench reached about 6.9x a single card. V-Ray GPU reached about 6x, and Redshift peaked at 4.5x. For an 8-card RTX 5090 machine, an iRender article states about 7.5x to 7.8x for Octane and 7.2x to 7.6x for Redshift, without published test details, so treat those as provider figures.

That matters when you choose a machine size. In the measured job above, the 4-card 5i machine cost $38 per hour. The 8-card 9i lists $62 per hour for runs under 3 hours, so it only saves money if eight cards finish this job at least 1.63 times faster than four. With near-linear Octane scaling, as in Puget’s OctaneBench result, eight cards would take about 41 minutes and cost about $58, roughly the same as the 5i. With Redshift-type scaling the 9i would cost more. Your engine’s scaling decides whether a bigger machine saves money, so render one test frame on 4 and on 8 cards before you budget the job.

6.3. What do measured jobs cost on a rented machine?

iRender publishes more real render times on its YouTube channel, each recorded on one of its machines, so those runs can be priced directly. The table below converts six of them at the listed hourly rate, without the 0.25-hour allowance used in table 6.1. It also shows what the same card-hours would cost at OB-hour rates, as a rough cross-check.

The measured jobs make the price gap concrete. A 300-frame Octane sequence on a 4x RTX 5090 machine cost about $0.17 per frame at iRender’s hourly rate, against roughly $0.09 to $0.12 for the same card-hours at OB-hour rates. In return, the job ran on a machine the user controlled directly, without a farm’s submission step, validation pass or job queue. The premium buys control, not raw price per card-hour. The video below records the Octane run from the first row.

6.4. The extra costs to add

6.5. A five-step estimate before you upload

1. Render three sample frames locally: one simple, one average and one heavy.
2. Note the frame time, the CPU or GPU, and your render engine version.
3. Convert the average frame time into each farm’s unit, using its own GHz or OctaneBench figures where published.
4. Apply the rate for the priority you need, then add tax, minimums and setup time.
5. Run the trial on two farms and compare the real per-frame cost with your estimate.

The RadarRender render farm cost calculator runs steps 3 and 4 for several farms at once. The render farm comparison page lists hardware, software support and the evidence behind each listing.

7. FAQ: render farm pricing

What is the difference between a GHz-hour and an OB-hour?

A GHz-hour measures CPU work: Super Renders Farm defines it as one CPU core running at 1 GHz for one hour, so the count is cores times clock speed times hours. An OB-hour measures GPU work against OctaneBench, OTOY’s benchmark, so a card that scores 1,730 uses 1,730 OB-hours in one hour. Both units let a farm charge for different hardware with one price list. Neither tells you how fast your own scene will render, which is why you still need a frame time from a test.

How do I calculate render farm cost per frame?

Render a few sample frames and note the time per frame and the hardware. Convert that time into the unit the farm bills: GHz-hours for CPU, OB-hours for GPU, or hours for a node or a rented machine. Multiply by the rate for the priority you plan to use. Then add tax, minimums, setup or idle time and any software or storage charges. Divide the total by the frame count. The RadarRender cost calculator runs these steps for several farms at once.

Why can two farms with similar rates produce different bills?

Rates hide three differences. Farms may count the work differently, and their pages rarely explain how a node’s GHz or benchmark figure is set. Hardware differs, so the same frame can take longer on an older node. Extras differ too, such as VAT, priority multipliers, RAM surcharges and whether queue or setup time is billed. Converting every quote to cost per frame with the same test scene is the only fair way to compare render farm pricing.

Are queue time, uploads and storage charged on a render farm?

It varies by farm, so read the pricing page. GarageFarm states that queue time, node acquisition and software installation are not charged. Super Renders Farm states there is no charge for uploading, queuing or downloading output. RebusFarm lists data storage, upload and download as free of charge. On machine-hour services the meter usually runs while the server is on, so time spent copying files or checking a scene costs money there.

Is renting a whole GPU server cheaper than a render farm?

Not by default. A rented server bills while it is switched on, including idle time, so short jobs lose more to overhead. iRender, for example, does not bill boot time and saves the machine environment between sessions. Its strengths are control and software coverage: you install your own tools once, and it can run real-time renderers such as Lumion, Enscape, D5 Render and Twinmotion, which job-queue farms generally do not support. For long batch renders in Corona, V-Ray or Cycles, a benchmark-billed farm may cost less. Compare both on cost per frame.

Do render farms offer free trials?

Several do, with limits. On 9 November 2026, GarageFarm listed $25 in trial credits, Super Renders Farm a $25 starter credit, RebusFarm 25 RenderPoints, Fox Renderfarm a $25 free trial, Ranch Computing a €30 trial offer for a first project and GridMarkets 5 free credits. AnimaRender offers a free demo render with watermarked output. iRender lists a 100% bonus on the first top-up, and its FAQ mentions a free trial on request through support. Use trial credit to render your sample frames, since that gives you the frame time every price estimate needs.

See more: Render Farm Comparison: Pricing, Hardware and Evidence

Sources

• GarageFarm, Pricing, read 9 November 2026.
• RebusFarm, Buy RenderPoints, read 9 November 2026.
• Super Renders Farm, Pricing and Pricing and credits, read 9 November 2026.
• Ranch Computing, Prices, read 9 November 2026.
• Fox Renderfarm, Pricing, read 9 November 2026.
• AnimaRender, Prices, read 9 November 2026.
• GridMarkets, Pricing, read 9 November 2026.
• Xesktop, GPU server types, read 9 November 2026.
• iRender, Price, FAQ and NDA article (irendering.net/is-your-3d-project-safe-on-a-render-farm-data-security-and-ndas/), read 9 November 2026.
• iRender, user documentation (Quick Start, Boot system, FAQ, Must-know tips, Reset machine, Pricing), read 10 November 2026.
• iRender YouTube channel, benchmark videos xCp-jHC0pPU, elkfXSepeTQ, zEeViBpOGvA, 6-LoVkQsy4w, PJ3eNgTMNKo, 9mvpps1TuWg and EuUroqBPANo (provider tests), read 10 November 2026.
• Puget Systems, 1-7x NVIDIA GeForce RTX 4090 GPU Scaling (November 2022), read 10 November 2026.
• iRender, Octane vs Redshift RTX 5090 multi-GPU scaling article (irendering.net/octane-vs-redshift-rtx-5090-multi-gpu-scaling/), provider test, read 10 November 2026.
• Lumion, knowledge base article on render nodes, read 9 November 2026.
• OTOY, OctaneBench 2025.2 results, read 9 November 2026.

The post Render Farm Pricing Explained: GHz-Hours, OB-Hours, Node-Hours and Per-Machine Billing appeared first on RadarRender.