Blender 5.1 render-farm benchmark
By Superluminal Engineering
A Blender 5.1 Classroom benchmark is being prepared across packaging, upload, queueing, rendering, delivery, and final-output checks.
Published 2026-07-29 · Updated 2026-07-29
Publication status
This page is a protocol, not a benchmark result. It is noindex because no complete set of repeated, end-to-end Blender 5.1 trials has passed the public evidence gate. There is currently no published median, p90 or p95, cost distribution, failure rate, output-equivalence result, or provider ranking.
The starting fixture is Classroom by Christophe Seux. Blender identifies its benchmark content as CC0, making Classroom suitable for a reusable public test while its creator and provenance remain visible. The target runtime entry is Blender 5.1.1, which is listed in Superluminal's audited runtime bundle. Listed does not mean the production-equivalent benchmark has already passed.
Evidence already available
A sanitized diagnostic report records one completed Blender 5.1 ZIP submission setup for Classroom:
| Field | Observed value | Status |
|---|---|---|
| Dependency records | 61 | Observed |
| Missing / unreadable | 0 / 0 | Observed |
| Files packed | 49 | Observed |
| Package bytes | 112,047,985 | Observed |
| Transferred bytes | 88,333,722 | Observed |
| Transfer elapsed | 16.326 seconds | Observed |
| Transfer errors | 0 | Observed |
| End-to-end publishable trials | 0 | Pending |
The report is useful setup evidence, but it does not measure queue wait, runtime preparation, scene load, render phases, output write, preview availability, download, validation, or repeated-run uncertainty. It cannot support a “fastest” claim.
Blender 5.1 benchmark progress
What is ready now and what still needs a complete Classroom run.
- Scene license
- CC0
- Blender build
- 5.1.1
- Public end-to-end trials
- 0
- Index status
- Noindex
Official Blender benchmark scene
Packaged runtime; full run coming
Coming after the full run
Hidden from search until results are ready
Benchmark boundary
Every trial must timestamp:
- dependency scan;
- package completion;
- upload start and finish;
- job registration;
- queue entry and assignment;
- runtime readiness;
- Blender startup or warm-session state;
- scene load and kernel preparation;
- frame render and output write;
- postprocessing and first preview;
- final output availability;
- download completion; and
- output validation.
The version inventory supplies the exact runtime label. The output-equivalence protocol defines how local and farm originals will be compared.
Trial design
The release must freeze the scene pack and checksum, Blender build, engine, device, settings, add-ons, dependencies, color configuration, frame set, output format, region, network conditions, cold or warm state, and pricing snapshot. Warm-up runs are labeled and excluded only under a preregistered rule. Failed and retried runs remain visible.
At least two scene classes are needed in addition to Classroom before making broad statements: one short-frame scene where overhead matters and one memory- or feature-heavy scene that exposes different bottlenecks. The result must report sample count, median, tail latency, range, exclusions, uncertainty, cost, and negative findings.
Required end-to-end timing boundary
- Local · Scan and package
Capture dependency scan, packaging duration, package bytes, and warnings.
- Network · Upload and register
Record region, network conditions, transfer duration, retries, and registration completion.
- Farm · Queue, environment, load, and render
Separate queue wait, runtime preparation, scene load, kernel work, frame rendering, writes, and retries.
- Output · Preview, download, and validate
Timestamp first preview, final availability, download, hashes, EXR checks, and declared tolerances.
- Release · Publish distributions and failures
Release sample count, median, p90 or p95, range, exclusions, failures, cost, raw bundle, and corrections path.
Release gate
This page can become indexable only when the raw bundle, analysis code, original outputs, sanitized logs, checksums, method version, reviewer, limitations, and corrections path are public and mutually consistent. Until then, the honest conclusion is simple: Superluminal has an exact Blender 5.1 target and a reusable CC0 fixture, but the end-to-end benchmark result is still being built.