Research
Blender GPU Runtime Reproducibility Under a Frozen Workload Contract
A frozen workload contract was validated for parity and control integrity. This report does not claim a completed V5 benchmark run.
Question
Can a frozen Blender workload contract stay comparable when only historical V4 evidence exists and V5 has not been physically executed?
Scope
Frozen V5 contract controls, not a V5 physical run. Historical V4 evidence is not V5 execution evidence.
- Contract
- FARPY-BLENDER-BENCHMARK-V5
- Version
- 5.0.0
- Status
- FROZEN
- Physical execution
- PENDING
- Runtime
- Blender 4.1.1
- Platform
- Linux x86_64
- Archive SHA-256
ab2ea3fe991601a5e6bd2cda786ecaa919c0b39e0550e59978b5d40270c260d3- Binary SHA-256
d5cf60563644ed770c5c13952624a0fa8ad076920bbc524275c0044029e36286
Result
Migration controls passed. No V5 performance result is published.
- Workload artifacts
- 8/8
- Frozen contract tests
- 10/10
- Scoring parity
- PASS
- Pass-count parity
- PASS
- CV-gate parity
- PASS
- Historical GPU
- NVIDIA GeForce RTX 3090
- Historical runtime
- 4.08s
- Historical score
- 24203
- Result ID
NMBENCH-V4-20260714T210332Z-898218- Configuration ID
NMBENCH-V4-202607
Historical V4 result only; not evidence of V5 execution
Method
The study checked artifact, scoring, and frozen-contract parity against the V5 contract. Physical V5 execution remains PENDING.
Limitations
- V5 has not been physically executed and no V5 performance result is published.
- No repeated runtime samples are asserted by this report.
- Evidence covers one verified GPU configuration and cannot establish hardware-wide reproducibility.
- Exact historical V4 runtime provenance remains insufficient.
Historical V4 evidence is not V5 execution evidence. This report publishes no V5 performance result.
What changed
FARPY treats V5 as a frozen contract, not a published performance result, until physical execution exists.
Evidence / files
blender-benchmark-v5-contract-20260719T031931Z