How Much Does It Really Cost to Render a 3D Animation? Local vs. Cloud Math
Source URL: Unsplash.com
Rendering a 720-frame animation on one RTX 4090 costs about $51. That is $2.71 of electricity at the EIA's April 2026 residential average, plus roughly $48 of depreciation. Published cloud render farm rates span a wide band per GPU hour. Your annual render volume decides which path wins.
Most artists learn what a render costs after the workstation has been pinned for two days. Every number you need is available before frame one, and the arithmetic takes about five minutes.
Local rendering hides its cost inside hardware you already paid for. That cost is calculable and varies depending on how much you render. Most published comparisons skip the second part, which is why their conclusions rarely transfer to your situation.
Which Five Inputs Set Your Render Bill?
Frame count and minutes per frame set your total machine time. Resolution, sample count, and render engine set the minutes per frame. Multiply the first two, and you have the whole job.
- Frame count scales the bill linearly. A 30-second clip at 24 fps runs 720 frames.
- Minutes per frame is your biggest lever, controlled through samples and scene cleanup.
- Resolution multiplies pixels. 4K carries roughly four times the work of 1080p.
- Sample count drives render time close to linearly before denoising claws part of it back.
- GPU engines such as Cycles or Redshift outpace CPU paths on most ray-traced work.
Per-frame times vary enormously between scenes. iRender's 2026 optimization write-up reports a heavy archviz frame at about four minutes, using 128 to 256 samples with the OptiX denoiser. The same frame at 2,048 samples without denoising took roughly 30 minutes.
Denoising is why that gap is so wide. It lets you cut sample counts hard without a visible drop in quality. Blender's adaptive sampling saves another 10% to 30% on top, per the Cycles developer documentation.
Frame weight gets decided upstream, during modeling and animation. A scene you cleaned at that stage renders cheaper at every stage after that. SelfCAD's walkthrough of the key steps from modeling to animation covers that groundwork.
Is Rendering On Your Own GPU Free?
No. Count electricity and hardware depreciation, and a single RTX 4090 workstation runs $1 to $5 per render hour. Where you land in that range depends entirely on how many hours you render each year.
How Much Power Does A Render Burn?
About 600 watts at the wall for a 4090 build. The bill is smaller than most artists fear, but it is not zero. Tom's Hardware rates the card at 450 watts of board power. Reviewer Jarred Walton measured it under sustained load in October 2022: "Both the 4090 and 3090 Ti average just under 440W in Metro Exodus." The rest of the machine pushes the total draw near 600 watts.
US residential electricity averaged 18.83 cents per kWh in April 2026, per the EIA's Electric Power Monthly. That works out to roughly 11 cents per render hour. UK readers pay more, with Ofgem's July to September 2026 cap set at 26.11 pence per kWh.
A 5090 build shifts these numbers. Its 575-watt board power pushes system draw somewhere between 750 and 900 watts under sustained load. Power still stays a minor line item for most artists.
What Does Depreciation Cost Per Render Hour?
Roughly $2 an hour on a typical build at moderate use, and it dwarfs the power bill. The figure is not fixed, and that is where most published comparisons go wrong.
Hardware loses value on a schedule. The IRS treats computer equipment as five-year property in Publication 946. Render farm operator SuperRenders puts the practical GPU resale decline near 25% a year, which on a $4,000 workstation is $1,000 of value gone annually. That happens whether the machine renders or idles.
Now divide that annual loss by your render hours:
- 200 render hours a year: $5.00 per hour
- 500 render hours a year: $2.00 per hour
- 1,200 render hours a year: $0.83 per hour
Same machine, same depreciation, three very different hourly costs. Add 11 cents of electricity, and you have your true local rate. Comparisons that quote one universal figure have skipped that division step.
One honest caveat on the 25% assumption. The 2026 GPU market broke the usual pattern. Used RTX 4090s traded near or above their launch prices after production ended and memory costs spiked. Treat 25% as a planning heuristic, not a market forecast.
What Does Rendering A 30-Second Animation Really Cost?
Source URL: Unsplash.com
Roughly $51 and a full day locally, against a wide cloud range and a few hours of wall clock time. The test job is 30 seconds at 1080p and 24 fps, so 720 frames. Assume two minutes per frame for a mid-complexity cycles scene.
That two-minute figure is an assumption, not a measurement. Time one frame from your own scene before you trust any of this arithmetic.
The Local Run: 24 Hours and About $51
720 frames at two minutes each is 1,440 minutes of unbroken rendering. Electricity comes to 14.4 kWh, or $2.71 at the EIA rate above. Hardware wear adds $48 at the $2 per hour figure.
Call it $51, plus a workstation you cannot touch for a day. You also carry the rerun risk alone. An out-of-memory crash at 3 am costs you the day and the power twice.
The Cloud Run: Hours Instead Of Days
Twenty-four GPU hours of compute lands somewhere between $25 and $200 across published farm rates. The spread comes from what you are renting. A bare GPU slot sits at the bottom, and a managed node with licenses, support, and priority queueing sits at the top.
Wall-clock time is where a farm wins reliably. Dozens of frames render in parallel, so a 24-hour job finishes in a few hours. You are usually buying time rather than savings. That matters most when a client deadline, rather than budget, is what constrains the job.
Comparing farms is harder than it should be. Rates arrive as node hours, GHz hours, OctaneBench hours, credits, and per-frame pricing. The only fair comparison is cost per finished frame on your own scene.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Which Costs Never Appear on the Price List?
The ones that decide most real projects. Both sides carry charges no pricing page shows you.
- Upload and download time on heavy scene files and 4K frame sequences
- VRAM ceilings, since multi-GPU setups do not pool memory for a single frame
- Heat and air conditioning load from 600 to 900 watts running in a small room
- Security requirements on NDA work, which rule out most cheap options
That last one is not negotiable on some jobs. Pre-release film and streaming work often requires vendors accredited under the Motion Picture Association's Trusted Partner Network. Farms holding both TPN accreditation and ISO 27001 certification, such as Fox Renderfarm, clear that bar. Community farms distribute your files to volunteer machines, so they fail it by design.
The VRAM point catches people out too. A 4090 gives you 24GB and a 5090 gives you 32GB per card. Adding a second card does not give you 48GB for one frame, because each GPU holds a full copy of the scene.
How Do You Find Your Own Break-Even Volume?
Divide your annual hardware depreciation by the gap between the cloud rate and your local power cost. The result is the render volume where owning starts beating renting.
Break-even hours per year = annual depreciation ÷ (cloud rate per hour − local power cost per hour)
Run it with the example numbers. $1,000 of depreciation, a $1 cloud rate, power at 11 cents. That gives about 1,120 render hours a year, or roughly 94 hours a month.
The answer swings hard with the cloud rate you would genuinely pay. At $3 per GPU hour, the break-even drops to around 350 hours a year. SuperRenders' 2026 analysis reaches similar ground from a different direction, treating 100 monthly render hours as roughly neutral.
The formula also explains why studios run both. Owned machines cover the baseline they can keep busy. Rented nodes absorb the spikes nobody sane buys hardware for.
When Does Local Rendering Still Win?
Source URL: Pexels.com
More often than farm marketing suggests, and the cases are easy to spot.
- Short jobs stay home, since upload and queue time swallow the gain on a ten-minute render
- Lookdev stays home, because live viewport feedback beats any farm while you dial lighting
- High-volume shops stay home past their break-even, keeping the farm for crunch weeks
There is a less obvious reason too. Your workstation does everything else: modeling, simulation, caching, and viewport work. A farm takes over one of those jobs and leaves the rest with you. SelfCAD's guide to the art of 3D rendering covers the lookdev stage in more depth.
Key Takeaways
- Local rendering costs $1 to $5 per hour once depreciation and electricity are counted.
- Your annual render volume, not any farm's rate card, decides which option is cheaper.
- Cloud rendering buys wall-clock time reliably and money only sometimes.
Conclusion:
Your local cost per render hour settles this argument, and almost nobody works it out. Time one frame, multiply by your frame count, and price both paths before the next job starts. Lookdev and short sequences belong on your own GPU. Long sequences belong on a farm once a locked workstation costs you more than the invoice would. Buying hardware is a bet on your own volume, so check the number before you place it.
FAQs:
How is render farm pricing calculated?
Farms bill in node hours, GHz hours, OctaneBench hours, credits, or per frame. Multiply the rate by the computer your job consumes. Managed farms such as RebusFarm and GarageFarm.NET bundle software licenses into the rate, while bare-slot providers do not. Compare cost per finished frame instead of headline hourly rates.
How long does a 5-minute animation take to render?
A 5-minute animation at 24 fps is 7,200 frames. At two minutes per frame on one RTX 4090, that is 240 hours, or ten straight days. A farm running frames in parallel compresses the same job into roughly a day, depending on queue load and priority tier.
How much electricity does GPU rendering use?
A 4090 render workstation draws about 600 watts under load, since the card alone is rated at 450 watts by Tom's Hardware. A 24-hour job burns 14.4 kWh. That comes to about $2.71 at the EIA's April 2026 US residential average of 18.83 cents per kWh.
Is the RTX 5090 faster than the RTX 4090 for Blender?
Yes. Blender Open Data lists a median score of 15,041.74 for the RTX 5090 across 3,061 submitted benchmarks, accessed July 2026. That leads the RTX 4090 by close to 30% on the same test. Faster hardware lowers per-frame time and pushes your break-even toward owning.
Do render farms offer free trial credits?
Many render farms offer trial credits, but the amount changes often. Use any available credit to test one of your own frames instead of relying on published benchmark averages.
About Bio:
Faisal Ali is the founder of GrowthBacklinks and a digital PR and SEO specialist who works with 3D, creative-tech, and SaaS brands. He turns technical topics into editorial content that ranks and earns links