Build vs Cloud Render Farm: The Real Cost Breakdown You Need Before Deciding
Overview
Introduction
Every year, studios face the same decision: build a local render farm or pay for cloud rendering? The answer seems simple. Run the numbers, pick the cheaper option. But most cost comparisons miss half the picture.
We have been operating a production render farm for over 15 years, supporting V-Ray, Corona, Redshift, Arnold and other render engines across the DCC applications we support. In that time, we have watched studios build farms, outgrow them, replace them, and sometimes abandon them entirely. The pattern is consistent: the upfront hardware purchase feels like the big decision, but it is the ongoing costs (render engine licenses, software maintenance, power, cooling, IT time) that determine whether building was actually worth it.
This guide breaks down the cost categories involved in building a local render farm in 2026 and compares them against cloud rendering. Every input is stated. Where a figure is our own planning estimate rather than a vendor quote, we say so and show the method, so you can replace it with your own number. The cloud side uses our billed rates, before any volume discount.
Whether you are a 5-person archviz studio rendering 50 hours a month or a 30-person VFX house pushing 500+, the math works differently. The goal here is to give you the framework to run your own numbers accurately, not to tell you what to do.
The True Cost of Building a Render Farm in 2026
Hardware costs depend on whether you are building a CPU-based farm (for V-Ray, Corona, Arnold CPU workflows) or a GPU farm (for Redshift, Octane, V-Ray GPU). All hardware prices in this section are our planning estimates, not vendor quotes. Check current listings before you budget.
CPU Render Farm (10 Nodes)
A production-grade CPU render node typically runs dual Intel Xeon processors. We price a 128 GB configuration, and we price the CPUs used: the Xeon E5-2699 V4 is a 2016-generation part. Per-node cost ranges from $1,900 to $2,900 depending on configuration. The CPU and motherboard line comes from listing prices we read on 8 November 2026 (used E5-2699 V4 chips at roughly $125-$155 each, used dual-socket LGA 2011-3 boards at roughly $165-$380), which gives $415-$690 for two chips and a board, rounded to $400-$700. Those listings are secondary sources, are not always dated and vary by seller, so treat the line as our estimate. A node built on current-generation CPUs costs more and renders faster per node; we have not priced one, so the CPU figures in this article apply to this older class of node.
| Component | Per Node | 10 Nodes |
|---|---|---|
| 2x used Xeon E5-2699 V4 (44 cores) + dual-socket board | $400-$700 | $4,000-$7,000 |
| 128 GB ECC RAM | $800-$1,200 | $8,000-$12,000 |
| 1 TB NVMe + case + PSU | $700-$1,000 | $7,000-$10,000 |
| Network switch + cabling | — | $500-$1,000 |
| Total | $1,900-$2,900 | $19,500-$30,000 |
GPU Render Farm (5 Nodes)
The NVIDIA RTX 5090 carries 32 GB of GDDR7 memory (NVIDIA product page, read 8 November 2026). We use a planning range of $2,500 to $3,800 per card, which sits above the card's launch list price. Street prices move, so check current listings. Custom and liquid-cooled variants cost more.
| Component | Per Node (1 GPU) | 5 Nodes |
|---|---|---|
| RTX 5090 (32 GB VRAM) | $2,500-$3,800 | $12,500-$19,000 |
| CPU + motherboard + 64 GB RAM | $2,000-$2,800 | $10,000-$14,000 |
| 1200W PSU + case + cooling | $500-$800 | $2,500-$4,000 |
| Network + shared storage | — | $3,000-$5,000 |
| Total | $5,000-$7,400 | $28,000-$42,000 |
For a dual-GPU-per-node configuration, common for Redshift and Octane production work, multiply GPU costs accordingly. A 5-node, 10-GPU farm lands between about $40,500 and $61,000 at these ranges.
These are day-one costs. Hardware starts losing value immediately. In every table below we spread the purchase price evenly over four years (25% of the price per year) and count no resale value. Newer GPU generations also raise the performance bar, so a farm built today will look slower next to what ships in a few years; we do not put a number on that.
The Licensing Trap
Hardware gets all the attention, but licensing is a separate cost line, and it is easy to leave out of a farm budget. How farm nodes are licensed differs by vendor and plan, so read the current terms before you budget. As one example, OTOY states that all Octane Studio+ subscriptions include 10+ render node licenses (read 8 November 2026).
This article does not put license prices in its tables. Seat prices differ by vendor, plan, billing period and region, and the vendors change them, so a figure printed here would age quickly. Read current prices at the source: Chaos pricing for V-Ray and Corona, Maxon pricing for Redshift and Cinema 4D, the Autodesk Arnold page for Arnold, and the OTOY purchase page for Octane. Two facts we confirmed at the source: Maxon states that Redshift is included with every Cinema 4D subscription (read 8 November 2026), and Blender, which includes Cycles, is free to use for any purpose (read 8 November 2026), so it needs no license fee.
To add your own license quote to our figures: every $100 per node per year of license cost raises the monthly break-even by about 1.7 card-hours for a GPU node, or about 7 machine-hours for a CPU node (derived from the rates and power costs in the sections below).
DCC application licensing is a separate line and depends on the application and your contract. Batch or command-line rendering can be licensed differently from running the interactive application, and Cinema 4D, After Effects and others each have their own rules for render nodes. Read each vendor's current license terms before you cost a farm.
Hidden Operational Costs
The costs below are real and easy to leave out of a first farm budget. The hours and rates are our planning assumptions, stated so you can swap in your own.
Electricity and Cooling
We use $0.1447 per kWh, the US average commercial retail price of electricity for August 2026 (14.47 cents, EIA Electric Power Monthly, read 8 November 2026). We assume a CPU node draws 500 W under load, a one-card GPU node draws 700-950 W, and cooling adds 30-40% on top of compute power.
- CPU node: 500 W x $0.1447 per kWh x 1.30-1.40 = $0.094-$0.101 per render hour
- GPU node: 700-950 W x $0.1447 per kWh x 1.30-1.40 = $0.13-$0.19 per render hour
- A CPU node rendering half of a year (4,380 hours) costs $412-$444 in electricity and cooling, so ten of them cost $4,120-$4,440
- A one-card GPU node at the same 50% costs $577-$843, so five of them cost $2,884-$4,215
Power scales with render hours, and it also applies to non-productive loads such as failed renders, test frames and software updates. The tables in this article count power for render hours only; nodes left powered on while idle would add to it.
IT Administration
Someone has to maintain the farm: software updates, driver patches, troubleshooting failed jobs, managing storage, handling license servers. This is not optional work.
We assume 5-10 hours per week of maintenance across a 10-node farm, at $50 per hour. That is a planning assumption, not survey data. It works out to $13,000-$26,000 per year (5-10 hours x 52 weeks x $50), or 0.5 to 1 hour per node per week. For studios without dedicated IT staff, this work falls on artists, which carries an even higher opportunity cost: a senior 3D artist spending time on render node troubleshooting instead of billable project work.
Version Management and Storage
Keeping identical software versions, plugins and assets synchronized across 10 nodes requires coordination. A V-Ray point release or a Forest Pack update needs testing on one node before rolling out to all, and incompatible versions across nodes produce inconsistent render output.
Fast shared storage (10 GbE NAS or better) is a further cost; we plan $3,000-$8,000 upfront plus ongoing maintenance. Without it, artists manually copy project files to each node, which is slow and error-prone.
Total Cost of Ownership: A Worked Example
Consider a 12-person architectural visualization studio rendering 100 V-Ray CPU machine-hours a month. We use this profile as a planning example. Both options below are priced at that same volume.
Option A: Build a 10-Node CPU Farm
| Cost Category | Annual Cost |
|---|---|
| Hardware (depreciated over 4 years) | $4,875-$7,500 |
| Render engine licenses (10 nodes) | Not priced here (see The Licensing Trap) |
| DCC licenses (if applicable) | Not priced here |
| Electricity and cooling (1,200 render hours at $0.094-$0.101) | $113-$122 |
| IT administration (8 hrs/week at $50) | $20,800 |
| Shared storage maintenance (planning allowance) | $1,500-$2,000 |
| Software troubleshooting and updates (planning allowance) | $2,000-$4,000 |
| Annual total, before licenses | $29,288-$34,422 |
Option B: Managed Cloud Rendering
| Cost Category | Annual Cost |
|---|---|
| Render hours (100 machine-hours/month x 12 x $1.2705) | $1,525/year |
| Volume discount (automatic on credit top-ups; not applied here) | $0 |
| Render engine licensing | Included in the rate |
| IT administration | Minimal |
| Hardware, electricity, cooling | None |
| Annual total, before volume discount | $1,525 |
The rate used is our billed price for a 72 vCPU machine at Standard priority, $1.27 per machine-hour; the table uses the unrounded $1.2705. For GPU work the billed rate is $5.17 per RTX 5090 card-hour. Credit top-ups get an automatic volume discount of up to 30%. The 30% tier requires a single top-up of 10,000 credits, so a studio spending about $1,500 a year will usually see a smaller discount, and the table shows the price before any discount.
At 100 machine-hours a month the two columns are far apart: about $1,525 for cloud against $29,288-$34,422 for the ten-node build, before licenses. Most of that gap comes from sizing. Ten nodes offer about 7,300 machine-hours a month, and this studio uses 100 of them (1.4%). The hardware depreciation and IT time are paid whether the nodes render or not. Even at 300 machine-hours a month, cloud would cost about $4,574 a year at list price.
A fairer comparison for this volume is a single node. One CPU node costs about $1,888-$3,447 a year here (depreciation, power and IT, before licenses), which is still above the $1,525 cloud bill at 100 machine-hours a month. The next section sets out that math for one to three nodes, and for ten.
For a detailed per-frame cost breakdown across different render engines and scene complexities, see our render farm cost per frame guide.
Break-Even for 1 to 10 Nodes
Studios often consider something smaller than a ten-node farm: one to three render nodes next to the workstations. We kept three cost categories from the example above (hardware depreciation, electricity and cooling, IT time), scaled them to one node, and added them up for one, two and three nodes. A ten-node CPU row is included so you can compare it with the example. For the solo-workstation case in more depth, see our single 5090 workstation vs cloud rendering cost analysis.
Inputs per node
| Input per node | GPU node (1 RTX 5090) | CPU node (used dual Xeon, 44 cores) |
|---|---|---|
| Hardware | $5,000-$7,400 | $1,900-$2,900 |
| Depreciation per year (4 years) | $1,250-$1,850 | $475-$725 |
| Electricity and cooling per render hour | $0.13-$0.19 | $0.094-$0.101 |
| IT administration per year | $1,300-$2,600 | $1,300-$2,600 |
Hardware comes from the per-node rows in the hardware tables above. Electricity and cooling per render hour comes from the power section. IT time takes the 5-10 hours per week planning assumption for 10 nodes and shares it evenly, which is 0.5 to 1 hour per node per week at $50 per hour. Render engine licenses are not in this table; see the assumptions below for how to add them.
Break-even by number of nodes
Break-even here is the fixed cost per year, divided by the cloud rate minus the electricity and cooling per hour the owned node would use, then divided by 12 months. The cloud rates are our billed rates before any volume discount: $5.166 per RTX 5090 card-hour for GPU, and $1.2705 per machine-hour for a 72 vCPU machine at Standard priority for CPU. A calculation with the rounded rates shown elsewhere in this article can come out an hour or two different.
| Nodes | Fixed cost per year, without IT | Fixed cost per year, with IT | Break-even per month, without IT | Break-even per month, with IT |
|---|---|---|---|---|
| 1 GPU node | $1,250-$1,850 | $2,550-$4,450 | 21-31 card-hours | 42-75 card-hours |
| 2 GPU nodes | $2,500-$3,700 | $5,100-$8,900 | 41-62 card-hours | 84-149 card-hours |
| 3 GPU nodes | $3,750-$5,550 | $7,650-$13,350 | 62-93 card-hours | 127-224 card-hours |
| 1 CPU node | $475-$725 | $1,775-$3,325 | 34-52 machine-hours | 126-237 machine-hours |
| 2 CPU nodes | $950-$1,450 | $3,550-$6,650 | 67-103 machine-hours | 251-474 machine-hours |
| 3 CPU nodes | $1,425-$2,175 | $5,325-$9,975 | 101-155 machine-hours | 377-711 machine-hours |
| 10 CPU nodes | $4,750-$7,250 | $17,750-$33,250 | 336-517 machine-hours | 1,257-2,370 machine-hours |
Monthly break-even per node against the billed cloud rate: used 44-core CPU node 126 to 237 machine-hours with IT time, one RTX 5090 node 42 to 75 card-hours with IT time
Hours are added up across the whole cluster: card-hours for the GPU rows, machine-hours for the CPU rows. Break-even scales in a straight line with the number of nodes, so the figure per node is the same in every row: a single RTX 5090 node breaks even against the cloud at 21-31 card-hours per month counting depreciation and power, and at 42-75 card-hours per month once IT time is included. For a used 44-core CPU node the same figures are 34-52 and 126-237 machine-hours per month. A month has about 730 hours, so a CPU node must be rendering roughly 17-32% of the time (IT included) to beat the cloud rate, and a GPU node roughly 6-10%. The CPU rows sit higher because the CPU rate per hour is lower than the GPU rate.
The ten-node row answers a question the example above raised: a 10-node CPU farm needs about 1,257-2,370 machine-hours of work every month, out of roughly 7,300 available, to break even with IT time counted.
What these figures assume
- Render engine licenses are not included. This article prices no licenses (see The Licensing Trap). Add your quote: each $100 per node per year adds about 1.7 card-hours (GPU) or about 7 machine-hours (CPU) to the monthly break-even. Blender, which includes Cycles, is free to use, so for Blender work these figures apply as shown.
- Three cost lines from the 10-node example are left out: DCC licenses (if applicable), shared storage maintenance, and software troubleshooting and updates. A real small farm carries some of them, so its cost runs higher than the table shows.
- Network and shared storage are not included. This article has no network or storage figure sized for 2 or 3 nodes. The prices it does give are for larger builds: switch and cabling at $500-$1,000 for the 10-node CPU farm, network and shared storage at $3,000-$5,000 for the 5-node GPU farm, and fast shared storage (10 GbE NAS or better) at $3,000-$8,000 upfront. A real 2 or 3 node cluster needs shared network and storage too.
- The CPU rows assume equal work per hour. They treat 1 machine-hour on our 72 vCPU machine as doing the work of 1 node-hour on the 44-core node used in this article. We have not measured that. Smaller workers bill less per hour ($0.99 and $1.17 for 48 and 56 vCPU machines), which would raise the CPU break-even. The GPU rows compare the same RTX 5090 card, so they do not depend on it.
- Break-even assumes that volume every month for all four years. The hardware is paid for on day one; a slow year does not refund it.
- The table compares money only. It puts no value on running a job in parallel across several machines, and none on reselling the hardware later.
- The cloud rate is before volume discount. The automatic discount on credit top-ups is up to 30%, and the 30% tier needs a single 10,000-credit top-up. At the break-even volumes above, a month of cloud spending is roughly $200 to $400 for one GPU node, so most users will earn a smaller discount. Any discount lowers the cloud rate and raises the hours needed to break even.
But I Already Own My Workstation
This is the most common objection we hear, and it deserves a direct answer.
If you already own render hardware, the purchase cost is a sunk cost. It is spent regardless of what you do next. The question is not "did I waste money?" but "what is the most productive use of this hardware going forward?"
Three factors matter:
Depreciation is ongoing. Hardware loses market value over time, and newer generations raise the performance bar. This article counts no resale value, so if you could sell an older card today, the sale price you give up by keeping it is a real cost of holding on.
Opportunity cost is real. Every hour your workstation spends rendering is an hour it cannot be used for modeling, texturing or scene setup. For a solo artist, this means choosing between productivity and rendering. Sending renders to cloud infrastructure keeps your workstation free for interactive work while renders run elsewhere.
Maintenance costs continue. Electricity, cooling, software updates and troubleshooting do not stop because the hardware is paid off. A "free" render node still costs power at the rates above, any engine licenses it needs, and the time spent managing it.
The most practical approach for studios with existing hardware: use your local machines for quick test renders and viewport previews, then send final production renders to cloud infrastructure. This hybrid model maximizes the value of hardware you already own while avoiding the scaling limitations and maintenance burden of a full local farm.
For solo freelancers and 1-2 person studios still deciding whether to buy a single high-end GPU like an RTX 5090 versus paying for cloud rendering per-hour, the math runs at a different scale than the studio TCO breakdown above. Our single 5090 workstation vs cloud rendering cost analysis walks through the indie-scale break-even threshold, opportunity cost for billable artists, and the specific scenarios where buying still makes sense at solo scale.
When Building Your Own Farm Actually Makes Sense
Local render farms are not always the wrong choice. They make economic sense under specific conditions:
- Consistent high volume per node: Each node you own has to stay busy. By the table above, a CPU node needs about 126-237 machine-hours of work a month (17-32% of the month) to beat the billed cloud rate with IT time counted, and a one-card GPU node needs about 42-75 card-hours. A 10-node CPU farm needs 1,257-2,370 machine-hours in total, every month, with minimal seasonal variation.
- Data security requirements: Compliance mandates (government contracts, NDA-heavy entertainment projects) require all data to remain on-premises with no external transfer. This is a hard constraint: if it applies, it can rule cloud out whatever the cost.
- Real-time iterative workflows: Rapid-fire test renders where network latency to a cloud farm would slow iteration cycles, though this typically applies to individual workstations, not farm-scale rendering.
- Existing IT infrastructure: You already have dedicated IT staff, rack space, power capacity and cooling, which reduces the marginal cost of adding render nodes.
Data security aside, the other three conditions are economic. If none of them applies, the numbers almost always favor cloud rendering. Break-even depends on how busy each node will be, and the break-even section above gives the figures. For a deeper look at how render farms process jobs and where cloud fits into production pipelines, see our technical guide to how render farms work.
Cloud Rendering: What You Actually Pay For
Not all cloud rendering services work the same way. The distinction that matters most is between IaaS (Infrastructure as a Service) platforms and fully managed farms.
We cover this distinction on its own in cloud-based rendering vs cloud computing rendering. It is worth a read before you total up either column below, since the two categories carry very different operational costs.
IaaS cloud rendering gives you remote access to hardware, essentially renting a machine. On generic IaaS platforms you usually install your own software, manage your own licenses and troubleshoot your own issues, though terms differ by provider. The hourly rate can be lower, but where you bring your own software you carry the operational work: license costs, software configuration, job management and debugging. How many hours that takes depends on your pipeline, so measure it for your own setup before comparing totals.
We also rent dedicated nodes, which is not the same as generic IaaS. Our render farm rental page lists dedicated GPU nodes (2x RTX 5090 per node) at $1,172.50 per node per week, which is $4,690 for a four-week month. That is a reduced reserved rate for committing to a full week, while the managed farm's rate is pay-as-you-go with no term. If a rented node rendered around the clock, that would be about $3.49 per card-hour, against $5.17 per card-hour on the managed farm. That is arithmetic on a full week, not an hourly price, and a rented node is a whole machine you connect to and run yourself, while the managed farm takes submitted jobs. The rental page splits licensing into two buckets. Licenses included: 3ds Max, Maya, Cinema 4D, Houdini and the render engines V-Ray, Corona, Arnold, Redshift and Octane are licensed at the node level, with the license included in the rental rate; Blender is in the stack too and needs no license fee. Bring your own license (BYOL): for everything else, you install your own licensed copy on the rented node. Licensing covers the stack, but not every engine and DCC pair is set up ahead of time: some pairs, such as Octane or V-Ray on Blender or Houdini, are provisioned on request, so contact us to confirm before you upload. The $3.49 against $5.17 comparison depends on a full-week reservation and on the node being busy all week; it is not a licensing difference. Rental suits teams with a steady load and the skills to run their own machines; the managed farm suits variable load and small teams.
Fully managed cloud rendering puts the supported render engines, job management and technical support inside the per-hour rate. The operational work that costs a local farm operator IT time (our planning assumption is $13,000-$26,000 per year for ten nodes) is done by the service's own team and priced into the rate.
On our farm, the V-Ray, Corona, Redshift, Arnold and Octane licenses are included in the rendering rate, so you do not buy or install render engines on farm machines. Jobs are submitted through web upload or the SuperRenders Client App, rendered on farm machines, and the output is downloaded. A lower hourly rate does not always mean a lower total: where licenses are not included, add them, along with admin time and billed setup time, before you compare. To understand the differences between managed and self-service models in detail, read our managed vs DIY cloud rendering comparison. For a closer look at how GPU workloads factor into rendering costs, see our cloud rendering overview.
Managed Cloud Farm vs Remote Desktop Rental: The Hidden Math
The previous section introduced the IaaS vs managed distinction. Standard pricing pages show the advertised hourly rate, not the effective hourly rate, and the gap depends on what happens during a typical remote desktop rendering session. The rate and times in the example below are illustrative, not a quote from any provider.
License Seat Consumption
On remote desktop providers where you bring your own software, your own DCC license occupies that machine. If your license allows only one active session at a time, one subscription seat running on a cloud VM means one fewer artist working locally. For a studio with five Maya seats, sending two to remote rendering machines leaves three available for interactive work, a 40% capacity reduction during render sessions. Check your vendor's terms for how your seats can be used on remote machines.
Providers that include licensing, such as managed cloud farms and our own rental for its licensed stack, handle this differently. On our managed farm, render engine licenses are bundled into the service and the farm operates its own license pool. Your local seats remain available for artists to continue modeling, texturing and scene setup while renders run on the farm's infrastructure.
Billed Idle Time
Remote desktop billing starts when the machine boots, not when rendering starts. A typical session includes non-rendering time that you still pay for. The durations below are illustrative:
- Scene upload: Transferring project files, textures and assets to the remote machine (about 5-20 minutes depending on project size and connection speed)
- Software configuration: Verifying render engine version, loading plugins, checking scene compatibility (about 5-15 minutes)
- Post-render download: Pulling finished frames back to your local storage (about 5-15 minutes)
On a managed farm, you submit a scene file through the farm's upload system. The farm handles file distribution, software configuration and job queuing internally. You are billed for actual render time: the clock starts when frames begin processing, not when your upload starts.
Setup Overhead Per Session
Most remote desktop platforms require verifying software versions, reinstalling plugins if they have been updated, and configuring render settings each session. The example session below carries 25 minutes of setup before rendering starts, all billed at the advertised hourly rate.
This overhead compounds with frequency. A studio running 20 render sessions per month at 25 minutes of setup each spends about 8.3 hours monthly on configuration alone, before a single frame renders.
Effective Hourly Rate: The Real Math
Here is how an illustrative $9 per hour advertised rate translates to the effective rate for a typical GPU rendering session on a remote desktop platform:
| Component | Time | Cost at $9/hr |
|---|---|---|
| Machine boot + software check | 10 min | $1.50 |
| Scene upload and configuration | 15 min | $2.25 |
| Actual rendering | 60 min | $9.00 |
| Post-render download | 5 min | $0.75 |
| Total session | 90 min | $13.50 |
Effective cost breakdown of an illustrative remote desktop rendering session: a $9 per hour advertised rate becomes $13.50 per rendered hour when boot, upload and download time are billed
Effective hourly rate for actual rendering: $13.50, which is 50% higher than the advertised $9/hour. And this calculation assumes everything works on the first attempt. Failed renders, plugin conflicts or version mismatches add more billed non-productive time.
Three Models Side by Side
The table compares one GPU workload, 100 RTX 5090 card-hours of rendering a month, across the three models.
| Factor | Build Local (1 GPU node) | Remote Desktop (IaaS), illustrative $9/hr | Managed Cloud Farm |
|---|---|---|---|
| Upfront hardware cost | $5,000-$7,400 | $0 | $0 |
| Monthly cost (100 GPU render hours) | $226-$390 with IT time; $117-$173 without (depreciation and power; engine licenses not priced) | $900-$1,350 (advertised $9/hr up to $13.50 effective per rendered hour) | About $517 (100 x $5.166, before volume discount) |
| Render engine licenses | You buy (not priced here) | Depends on the provider: often yours, sometimes included; check the provider | Included in the rate |
| DCC license impact | No impact (dedicated node) | May consume your seats | No impact on your local seats |
| IT overhead per month | About 2-4 hours (0.5-1 hour per week, our assumption) | Depends on the provider: software, licenses and job management are yours where not included | Minimal (submit and download) |
| Setup time per session | One-time, then ongoing maintenance | 25 min in the example above | None per session for software that is ready on the farm; engines and plugins listed as on request are arranged with support before the first job |
| Hardware control | Full | Partial (provider's specs) | None (provider manages) |
| Scalability | Buy more hardware | Rent more VMs | Capacity comes from the farm's pool |
| Suited for | Volumes above the per-node break-even (42-75 card-hours a month with IT), or data that must stay on-premises | Teams with DevOps skills, custom pipeline requirements, large node counts | Small teams without IT staff, variable rendering volume |
At 100 card-hours a month, an owned GPU node costs less per month than the managed rate, even with IT time counted. That is consistent with the 42-75 card-hour break-even above. Below that volume the managed farm is cheaper; above it the owned node is. The right choice depends on your volume, team size and technical capacity. There is no universally correct answer.
For studios that have decided on cloud rendering and need to compare specific services, our render farm evaluation guide provides a structured checklist covering eight key criteria.
Decision Framework
| Factor | Local Farm Favored | Cloud Rendering Favored |
|---|---|---|
| Monthly render hours per node | Above about 126-237 machine-hours (CPU, used 44-core nodes) or 42-75 card-hours (GPU) per node, every month | Below that, or variable |
| Who maintains the nodes | IT or pipeline staff whose time is already budgeted | Artists would end up doing it |
| Utilization pattern | Steady, predictable year-round | Seasonal, project-based |
| Data requirements | Air-gapped, on-premises mandated | Standard NDA sufficient |
| Budget model | Large CapEx acceptable | Prefer OpEx, pay-per-use |
| Hardware refresh | Accept 3-4 year depreciation cycles | Want current-generation hardware |
| Pipeline staffing | Dedicated TD or IT admin available | No pipeline engineer on team |
| Render engine licensing | Already hold licenses that cover farm nodes | Prefer licensing included in the rate |
A hybrid model can work when your volume sits near or below the per-node break-even, or varies by season: local nodes handle iterative test renders while cloud infrastructure scales for final production deadlines. For more on evaluating cloud render farm options, see our guide to choosing a cloud render farm.
To understand how different cloud render farms charge, and how to calculate effective cost versus advertised rates across six pricing models, see our render farm pricing models comparison.
Often-Missed Costs: Summary Table (10-Node CPU Farm)
| Cost Category | Annual Range | Applies To |
|---|---|---|
| Render engine licenses | Vendor pricing; not priced here | Local + generic IaaS |
| DCC licenses (render nodes) | Vendor terms; not priced here | Local + generic IaaS |
| Electricity and cooling | $4,120-$4,440 at 50% utilization; $113-$122 at the 100-hour example | Local only |
| IT administration labor (planning assumption) | $13,000-$26,000 | Local + IaaS |
| Shared storage maintenance and troubleshooting (planning allowances) | $3,500-$6,000 | Local + IaaS |
| Hardware depreciation | $4,875-$7,500 (4-year straight line, 25% of price a year) | Local only |
Total annual cost for the 10-node local farm in the 100-hour example: $29,288-$34,422 before license fees (IT taken at 8 hours a week, $20,800), with IT labor and hardware depreciation as the two largest line items. Cloud rendering at the same usage (100 machine-hours a month) is about $1,525 a year at list price on a fully managed platform. The comparison is lopsided because the ten nodes sit mostly idle at that volume; see the break-even section for the volumes at which owning wins.
Current rates are on the pricing page. Credit top-ups get an automatic volume discount of up to 30%, with no code needed; the 30% tier needs a single top-up of 10,000 credits.
FAQ
Q: How much does it cost to build a 10-node render farm in 2026? A: A 10-node CPU farm (used dual Xeon, 128 GB RAM per node) costs about $19,500-$30,000 in hardware at our planning estimates. A 5-node GPU farm with RTX 5090 cards costs about $28,000-$42,000. These figures do not include render engine licenses (priced by the vendors), electricity ($0.094-$0.101 per CPU node render hour in our assumptions) or IT labor ($13,000-$26,000 a year at 5-10 hours a week and $50 an hour).
Q: What is the biggest hidden cost of running a local render farm? A: IT administration time. We assume 5-10 hours per week across a 10-node farm for software updates, driver patches, job troubleshooting, storage management and license server maintenance. At $50/hour, that is $13,000-$26,000 annually, which is more than the electricity for ten CPU nodes rendering half the year ($4,120-$4,440). It is a planning assumption, so replace it with your own hours.
Q: At what point does building a local render farm break even versus cloud rendering? A: It depends on how busy each node is. Against our billed rates before volume discount, one used 44-core CPU node breaks even at about 126-237 machine-hours per month with IT time counted (34-52 without), and one RTX 5090 node at about 42-75 card-hours per month (21-31 without). A 10-node CPU farm needs about 1,257-2,370 machine-hours per month in total. These figures assume four-year depreciation, exclude render engine licenses, and need that volume every month. Below it, cloud rendering costs less.
Q: Do render engines require separate licenses for each farm node? A: It depends on the engine and the plan, because farm-node licensing differs by vendor. For example, OTOY states that Octane Studio+ subscriptions include 10+ render node licenses. Blender, which includes Cycles, is free to use for any purpose, so it needs no license fee. Read each vendor's current terms before you budget; this article does not print per-seat prices, because they differ by vendor, plan and region.
Q: Is cloud rendering more expensive than using my own GPU workstation? A: It depends on volume. A single RTX 5090 node costs about $5,000-$7,400 to build. Against the billed cloud rate of $5.17 per card-hour (before volume discount), owning breaks even at about 42-75 card-hours a month once IT time is counted, or 21-31 card-hours counting depreciation and power only (engine licenses not included). Below that, cloud is cheaper; above it, owning is. At 50 card-hours a month the cloud bill is about $3,100 a year. At 200 card-hours it is about $12,398 a year, against roughly $2,866-$4,912 a year for an owned node with IT time.
Q: What is the difference between IaaS cloud rendering and fully managed cloud rendering? A: Generic IaaS gives you access to remote hardware, and you usually install software, manage licenses and troubleshoot issues yourself, though terms differ by provider. Fully managed farms include the supported software, render engine licenses and technical support in the per-hour rate. The hourly rate for IaaS can look lower, but where you bring your own software you take on the licensing, configuration and debugging work, and the billed setup time. Compare totals, not hourly rates.
Q: How fast does render farm hardware depreciate? A: In this article we spread the purchase price evenly over four years, which is 25% of the price a year, and we count no resale value. Actual resale prices vary with the market. Newer GPU generations also raise the performance bar, so hardware loses ground against current alternatives over a 3-4 year life; we do not put a throughput ratio on it.
Q: Can I use a hybrid approach, with local hardware plus cloud rendering? A: Yes, and it is a practical model. Use local workstations for quick test renders and viewport previews, then send final production frames to a cloud farm. This approach keeps your workstation available for interactive work, avoids the capital expense of building a full farm, and lets you scale rendering capacity for deadlines without maintaining idle hardware year-round.
Q: Is renting a remote desktop cheaper than a managed render farm? A: Not always. In our illustrative example, a $9/hour remote desktop session includes billed idle time for scene upload, software setup and post-render download, turning a 60-minute render into a 90-minute billed session at a $13.50 effective hourly rate. Managed farms bill for actual render time, include render engine licenses and need no per-session software setup for software that is ready on the farm (some engines and plugins are set up on request). Compare your own session times and license costs; at 100 GPU card-hours a month the example above comes to $900-$1,350 for the remote desktop and about $517 for the managed farm at list price.
Q: Do I need my own software licenses to render on a cloud render farm? A: It depends on the type of cloud rendering service. On a generic remote desktop or IaaS platform, usually yes: your own DCC and render engine licenses occupy the remote machine, which may reduce the seats available for local artists. On our farm rental, the licensed stack (3ds Max, Maya, Cinema 4D, Houdini, V-Ray, Corona, Arnold, Redshift, Octane) is licensed at the node level and included in the rental rate, Blender is in the stack and needs no license fee, and you bring your own license only for other software. Some engine and DCC pairs are provisioned on request, so contact us to confirm before you upload. On a fully managed cloud render farm, the render engine licenses are included in the rate. Check the farm's software support page for the applications and versions it covers. Your local licenses remain available for interactive work while the farm renders.
About Alice Harper
Blender and V-Ray specialist. Passionate about optimizing render workflows, sharing tips, and educating the 3D community to achieve photorealistic results faster.
