Skip to main content
NexGPU

Graphics rendering

Ten in parallel is a render farm

Rendering is the workload hourly rental was made for: it has a clear start and end, it parallelises perfectly, and it needs exactly zero compute when you are not rendering.

Sign up freeView pricing
RTX 4090 per hour
$0.540
Parallel means ten times faster
10x
Cost while idle
0

01 — When it applies

Why rendering belongs in the cloud

Ten seconds of animation is 240 frames, and every frame is independent of every other. One machine rendering 240 frames and ten machines rendering 24 each consume identical total compute — but the second finishes in a tenth of the time. That perfect parallelism is the entire reason render farms exist.

Hourly rental drops the barrier to owning one to zero. You do not need ten machines; you need ten machines on the night before delivery, and then you destroy them. Total spend equals one machine running ten times longer, and you get the result ten times sooner.

For solo artists and small studios that difference often decides whether a job is winnable at all. Project volumes that used to require a serious hardware budget can now be taken on by renting per project — and the cost drops straight into the quote.

02 — Workloads

Shapes of rendering work

From a single frame to a full sequence, product visualisation to architectural work.

  • Animation frame sequences

    Split the frame range across instances and merge the output. This is where cloud rendering pays most — wall-clock time compresses almost linearly with budget.

  • Product and architectural visualisation

    High-resolution stills, multiple angles, competing material treatments. Where a single frame takes a long time, a 4090 is several times faster than a mid-range local card.

  • Lighting and material iteration

    Tuning light and materials means repeated test renders. Keeping one instance running while you iterate beats watching a progress bar locally.

  • AI-assisted post

    Upscaling, denoising and frame interpolation after the render also need a GPU, and can run on the same instance — no shuttling files back and forth.

03 — Stack

Renderers and how to drive them

Command-line batch rendering is the most reliable way to use cloud GPUs.

  • Blender — the open-source default

    Cycles performs well on NVIDIA through the CUDA and OptiX backends. Batch render with blender -b and use frame-range flags to split work across instances directly.

    Blender · Cycles · OptiX · CLI batch

  • Commercial renderers

    V-Ray, OctaneRender and Redshift all install and run inside an instance. Note that you must hold the licences yourself — we supply compute, not commercial software licensing.

    V-Ray · Octane · Redshift

  • Remote desktop for scene work

    Run a remote desktop when you need a GUI to adjust the scene. Keep the rendering itself on the command line, though — it is steadier, lighter and far easier to script.

    Remote desktop · Visual debugging

04 — Choosing a GPU

Sizing a card for rendering

Rendering needs both VRAM and compute: VRAM decides whether the scene fits, compute decides how fast it finishes. A scene that does not fit fails outright, so leave headroom.

Scene sizeVRAM neededCheapest availableNotes
Small to mid scenes, product shots12GBRTX 3060$0.100/hrFine for typical product and small interior scenes. The best-value entry tier.
Complex scenes, high-res stills24GBTesla V100$0.188/hrHeavy geometry and high-resolution textures, where 24GB is the speed-per-dollar optimum.
Very large scenes, film-grade assets48GBQ RTX 8000$0.508/hrA scene that exceeds VRAM fails the render outright, so go straight to the 48GB tier for large asset libraries.
Multi-GPU render nodes80GBA100 SXM4$1.088/hrOne machine can render different frames on each card, or pool VRAM where the renderer supports it.

"Cheapest available" is derived from live inventory — the lowest per-GPU rate among models that clear the VRAM bar — and moves with the market. Multi-GPU nodes rent whole.

05 — Getting started

Running a cloud render

  • 01

    Pack the scene and assets

    In Blender use Pack Resources to embed textures into the .blend, or transfer the whole asset directory. Missing assets are the most common cause of a failed cloud render.

  • 02

    Boot N instances, one frame range each

    Work backwards from your deadline to decide how many. Give each instance its own -s/-e range so they never collide. Size the disk for the output sequence.

  • 03

    Collect the output and destroy them all

    Download the frames and composite. Once you have the files, destroy every instance — ten stopped instances means ten storage charges.

06 — FAQ

Cloud rendering

What does rendering ten seconds of animation cost?

It depends on per-frame time. If a frame takes one minute on a 4090, 240 frames is four machine-hours — about $2.16 at $0.540/hr. Ten instances in parallel costs the same total and delivers in 24 minutes.

What about commercial renderer licences?

You must hold them yourself and use them under their own terms. We provide GPU compute and an operating system, not commercial software licensing. Blender is open source, so the question does not arise there.

How do I transfer large scene files?

Pack first, then transfer. Up to a few GB, scp directly; beyond that, stage in object storage and pull from inside the instance over the host's bandwidth. This matters most when booting many instances at once, or file transfer becomes the bottleneck.

Will parallel machines produce colour differences?

No — with the same renderer version and the same scene file, GPU rendering is deterministic. Verify that every instance runs the same renderer version; using one image for all of them takes care of it.

What if a machine fails mid-render?

That is the advantage of splitting by range: only that segment is lost, and one replacement instance re-renders it without touching the others. Output image sequences rather than video directly, so any single frame can be redone.

Related solutions

Other ways to use it

Same compute network — swap the image and it becomes a different production line.

Start building on NexGPU

Enterprise R&D team or solo developer — either way, your first job can be running in minutes.

Sign up freeTalk to sales

Sign up to browse live network pricing. No payment method required.