Blender Cycles · Multi-GPU · EU

Your finished scene. Rendered while you keep working.

A small, attended GPU render setup in Poland for studios and freelancers. You send the prepared Blender scene, I render a free test frame, quote a fixed price and deliver the frames.

Frame rendered on this farm: the Blender demo scene Classroom
Rendered on this farm · Blender demo scene “Classroom” (Christophe Seux, CC0)

What I do, and what I don't

I take over the compute, not the design. Your scene stays your scene.

I render

  • Animations as image sequences, distributed across several GPUs
  • High-resolution stills, e.g. architectural and product views
  • Variants of the same scene: materials, cameras, times of day
  • Recurring jobs with a fixed Blender version and add-ons

Not offered

  • Modelling, lighting, look development or creative work
  • Render engines other than Cycles without prior testing
  • Rush jobs with thousands of frames in a few hours: a large cloud farm is the better choice for that

Hardware

Three Linux machines on the same network, managed with Flamenco, the Blender project's render manager. Every scene is tested on exactly this hardware first.

Close-up inside Node A: the two RTX 4090 graphics cards
Node A · Main workstation 2 × RTX 4090 · 24 GB VRAM each

Main capacity, including memory-heavy scenes.

Inside Node B: workstation with RTX 4080 SUPER
Node B · Workstation RTX 4080 SUPER · 16 GB VRAM

Additional worker for distributed frames.

Node C: MSI Vector A16 HX laptop with RTX 5070 Ti
Node C · Supplementary RTX 5070 Ti Laptop · 12 GB VRAM

Added for scenes that fit.

GPU memory is not pooled: each GPU renders with its own copy of the scene. Large scenes run on the 24 GB cards; animations benefit from all GPUs at once.

Measured render performance

Measured with the official Blender Benchmark: Blender 5.2, OptiX, each GPU on its own. Reference values are the median of all results in the public Blender Open Data database.

Blender Benchmark 5.2

as fast as 3.4 × a single RTX 4090

38,922 samples per minute, sum of the three benchmark scenes across all four GPUs. A benchmark value, not a guaranteed factor for every job: the test frame from your scene shows the real speed.

Our cards compared with the median

Cardmeasuredmediandifference
RTX 4090 #112,39911,542+7 %
RTX 4090 #212,10911,542+5 %
RTX 4080 SUPER8,7528,243+6 %
RTX 5070 Ti Laptop5,6625,110+11 %

Compared with single graphics cards

Thiry Origin · 4 GPUs38,922
RTX 509016,120 · 2.4 × faster
RTX 409011,542 · 3.4 × faster
RTX 4080 SUPER8,243 · 4.7 × faster
RTX 40705,106 · 7.6 × faster
RTX 30601,986 · 20 × faster
Ryzen 9 9950X (CPU)666 · 58 × faster
Rendered frame of the Blender demo scene Classroom
Frame 72 from the test. Scene: Christophe Seux, CC0, blender.org

Real-world test: the “Classroom” scene

145 frames at 1920 × 1080, spread across all three machines in blocks of five frames: done in 7 minutes 51 seconds, with no errors. About 90 Wh for GPUs and CPUs, roughly 0.6 Wh per frame. Under sustained load the GPUs stayed at 66 °C or below.

GPU temperature during the test (°C)02550751000 min2 min4 min6 min8 minRTX 5070 Ti LaptopRTX 4080 SUPERRTX 4090 #1RTX 4090 #2
Temperature per GPU over the whole run, peak value per sample window

Render it yourself or have it rendered?

Work through your project: time and cost on your own machine compared with the farm.

Your project
Your machine
For the time your machine is tied up.
Render it yourself –
Electricity
–
Machine tied up
–

Cost –

With Thiry Origin –
Compute
–
Price
–

Fixed price approx. –

Compute is billed in RTX 4090 hours: €1.50 per RTX 4090 hour, minimum order €10. All prices are final prices. For example, work that takes a single RTX 4090 ten hours costs €15, and the farm finishes it in about three hours. The estimate is based on benchmark ratios and pure render time; price and delivery date are fixed after the free test frame.

Measured on 10 Nov 2026 · Reference values: opendata.blender.org (CC0), as of 10 Nov 2026, Blender 5.2.0, OptiX, median per card. Totals = monster + junkshop + classroom.

Process

No account, no upload portal, no surprises on the price.

  1. RequestBlender version, frame range, resolution, output format and deadline by email.
  2. Hand-overPacked .blend file with all assets via an upload link to my self-hosted Nextcloud, encrypted over HTTPS.
  3. Free test frameI render one frame from your scene and measure the time. The test frame is always free.
  4. Fixed priceYou get price and delivery date before rendering starts.
  5. DeliveryFrames as EXR or PNG via download link from the same Nextcloud, with a render log.

Live project tracking

Your render job. Your progress.

Follow your render job through a private, read-only dashboard. See completed frames, current status and estimated completion time — without needing to ask me for updates.

  1. Free test frame & fixed price
  2. Live progress
  3. Finished frames

How your data is handled

Processing

  • Rendering only on my own hardware in Poland (EU)
  • File transfer through a self-hosted Nextcloud on a server in Germany, no third-party cloud services
  • Each project in its own separate render environment
  • Automatic execution of Python scripts in .blend files is disabled

After delivery

  • Project files are deleted 14 days after delivery, on the render machines and in the Nextcloud, unless agreed otherwise
  • For recurring projects, separate long-term storage on request
  • Non-disclosure agreement (NDA) on request

Request a free test frame

Write a few lines about the job. You'll get a reply with questions or a date for the test frame.

hello@thiryorigin.com

Useful in the first email

  • Blender version and add-ons used
  • Still or animation, number of frames
  • Resolution, samples, output format
  • Desired delivery date
  • One-off project or recurring need