What is a render farm?

How does a render farm work?

Rendering an animation is an easy job to share out, because each frame can be computed on its own. A render farm exploits that. The details vary between farms, but the chain is always roughly the same.

1. The scene and its dependencies are gathered. A render node only sees what it receives: the scene file, plus every texture, HDRI, linked library, font, simulation cache and plug-in the scene relies on. A texture referenced by an absolute path such as C:\Users\me\textures\wood.png does not exist on the farm, and the frame renders with a pink or missing texture. In Blender, relative paths and File > External Data > Pack Resources avoid most of these surprises; baked caches still have to travel with the file.
2. The job is submitted. On a cloud farm, this means uploading the scene through a web page or a plug-in. On a local farm, the file sits on shared storage that every machine can read. The job carries its settings: frame range, output format, software and renderer version.
3. A render manager queues it. The manager is the software at the centre of the farm. It splits the job into tasks, usually one frame or a small chunk of frames each, keeps a queue ordered by priority, hands each task to the next free node, and sends a task to another node when one fails. Deadline, OpenCue and, for Blender, Flamenco are examples of render managers.
4. Each node renders its frames. A node opens the scene, prepares it (loading textures, building the acceleration structure, compiling shaders) and renders. That preparation is repeated on every machine, so it weighs more when frames are short.
5. The output is collected. Finished frames are written to shared storage or sent back to you, usually as an image sequence. You then assemble them into a video or pass them to compositing.

The speed-up follows from the arithmetic. With 250 frames and 10 machines, each machine renders 25 frames, so the job takes about a tenth of the time, plus the preparation of each node and the transfer of the files. A farm cannot use more machines than there are frames: a single still image goes to one machine. Some farms split a still into tiles or into sample passes rendered on several machines and merged afterwards, which helps only for very long renders.

CPU vs GPU render farms

The nodes of a render farm compute either on their processors or on their graphics cards, and the choice depends on the renderer the scene uses.

• CPU render farms suit renderers that run on the processor, such as Corona or the CPU modes of Arnold and V-Ray. Their strength is memory: a CPU node can hold very large scenes in system RAM, which matters for heavy VFX and archviz.
• GPU render farms suit GPU renderers such as Redshift, Octane, and Cycles with OptiX, CUDA or HIP. For Cycles, the Blender Open Data benchmark puts recent graphics cards well ahead of desktop processors. The limit is video memory: the scene has to fit in the card's VRAM, and a scene that does not fit fails or slows down sharply.

Blender's Cycles runs on both, and Eevee needs a GPU. Before choosing a GPU farm, check that your scene fits in the memory of the cards it uses: the render window in Blender reports the memory used by a test frame.

Cloud rendering vs building your own render farm

Cloud rendering means rendering on machines rented over the internet rather than on your own. A cloud render farm is the most common form: you upload the scene, and remote machines render it. The alternative is to build a farm from computers you own, which is what studios with a steady flow of renders often do. The comparison is less about the price of a GPU than about everything around it.

• Hardware. Each node needs its own GPU or CPU, memory, power supply and storage, plus a network switch and shared storage fast enough to feed every node at once. Hardware loses value whether it renders or not.
• Electricity and heat. A workstation with a high-end GPU draws several hundred watts under full load. The running cost is the power at the wall, times the hours of rendering, times your electricity rate, and every watt ends up as heat that the room has to get rid of. Several nodes in an office also make noise.
• Licences. Blender is free on every machine. Commercial renderers and plug-ins usually need a render node licence per machine, a recurring cost to add to the hardware.
• Maintenance. Every node needs the same software version, the same add-ons and compatible drivers. Someone has to update them, replace failed parts, and find out why frame 143 renders differently on one machine.
• Utilisation. This decides the outcome. A farm that renders most days pays for itself over time. A farm used for one deadline a month spends most of its life idle, while a cloud farm is paid only for the hours used.

The cloud has its own costs: uploading large scenes takes time, the files leave your machine, and a farm only renders the software and versions it supports. A local farm has no upload and no dependency on a provider.

Are render farms worth it?

A render farm buys time. Whether that is worth paying for depends on how much time is at stake and what it is worth to you.

Usually worth it when:

• you render an animation, where hundreds of frames can be spread across machines;
• the render on your computer would take longer than the time left before the deadline;
• you need your computer for other work while the render runs;
• you render rarely enough that owning a farm would leave it idle.

Usually not worth it when:

• you render a single still that takes a few minutes locally: uploading the scene takes longer than the render, and one frame cannot be split between machines;
• you are still adjusting lighting and materials, since test renders are faster on your own machine;
• the scene relies on a plug-in or a version the farm does not support;
• you have no budget and no deadline. A free community farm such as SheepIt, or simply rendering overnight, will do.

Before paying for any farm, reduce the time per frame: it multiplies the cost of every frame. The guide to rendering faster in Cycles and the guide to choosing the noise threshold cover the settings that matter.

How much does a render farm cost?

Render farms do not share a billing unit, so prices can only be compared on your own scene. The common models:

• GHz-hour, mostly on CPU farms: number of cores, times clock speed in GHz, times hours. A 32-core node at 3 GHz running for one hour counts as 96 GHz-hours.
• Node-hour or GPU-hour: the time a machine or a card spends on your job, at an hourly rate. Some GPU farms use a benchmark unit instead, such as an OctaneBench score multiplied by hours.
• Per frame: a set price per rendered frame, sometimes varying with resolution or render time.

Whatever the model, also check the minimum charge per job, the priority levels (faster often costs more per hour), fees for storage or transfer, and whether renderer licences are included. Then run a test: render one representative frame, note its time, and multiply by the number of frames. The render time calculator does that arithmetic for you.

Here is the same calculation with TURNA's public tiers, for 250 frames at 2 minutes per frame, a 10-second shot at 25 fps. On one computer, that is 8 h 20 of rendering. The estimate assumes each GPU renders a frame as fast as the machine where you timed it.

• Standard, 1 GPU at €2.90 per hour: about 8 h 20, €24.17.
• Boost, 2 GPUs at €7.90 per hour: about 4 h 10, €32.92.
• Max, 4 GPUs at €19.90 per hour: about 2 h 05, €41.46.
• Turbo, 8 GPUs at €49.90 per hour: about 1 h 03, €51.98.

More machines finish sooner and cost more in total, because the hourly price of each tier includes a premium for speed. The add-on shows the estimate for your own scene before anything starts.

What types of render farms are there?

• Managed render farms. You upload the scene, the service runs the software and the queue, and you download the frames. GarageFarm, RebusFarm and Ranch Computing are examples. Check which applications, renderers and versions each one supports.
• Peer-to-peer render farms. Members' computers render each other's frames. SheepIt is one, for Blender only: it is free, and you earn points by letting your machine render other people's projects. Waiting times depend on how many machines are online and how busy the queue is.
• DIY render farms. Your own computers, linked by a render manager such as Flamenco for Blender. Full control and no upload, with the costs described above.
• Infrastructure as a service (IaaS). You rent remote machines by the hour, from general cloud providers or from GPU rental services such as iRender, and install and run the software yourself. More flexible than a managed farm, with more setup and administration.

Frequently asked questions

Rendering the animation on TURNA

When a scene is already well optimised and the frame count is what takes the time, the remaining lever is rendering several frames at once. TURNA does this for Blender: an add-on (Blender 4.0 to 5.2, Cycles and Eevee) sends the scene in one click, and the frames are split across 1, 2, 4 or 8 GPUs in parallel, RTX 5090 first and RTX 4090 when none is free. The add-on shows the estimated cost before anything starts. Prices start at €2.90 per hour, billed on the time actually used, with a 30-second minimum, and new accounts get €25 of free credit. A single still image is still rendered by one machine, so the gain is on animations. Tiers and worked examples are on the Blender render farm page.

In the terms used above, TURNA is a managed render farm for Blender only. GPUs are looked for in Europe first; the exact rule, and what happens when none is free there, is on the Blender render farm page. The four tiers and the example worked out for each of them are on the pricing page.