Benchmark measurement details¶
Historical measurement. Not a measurement of the latest release.
View the results · Contributor benchmark guide
| Measurement identity | Value |
|---|---|
| Source revision | 5db7197532eb08b37d748a78fb789545ab8aa577 |
| Measured at | 2026-10-21T11:26:29.269699Z |
| Result status | completed |
| Source document SHA-256 | 0b77be28e24a2c6fb6fab834b1e4dc27deaf20714b782fd4bb24178f5e566c6d |
| Policy | Paired workload comparisons are summarized within each runner cohort; runner cohorts are not compared against one another. |
| runner_count | 4 |
| platforms | macOS 15 · arm64; Ubuntu 24.04 · arm64; Ubuntu 24.04 · x86_64; Ubuntu 24.04 · x86_64 · Pillow-SIMD paired cohort |
Runner cohorts and hardware¶
| Runner cohort | OS | Architecture | CPU | Python | Rust toolchain | Runner image |
|---|---|---|---|---|---|---|
| macOS 15 · arm64 | macOS-15.7.9-arm64-arm-64bit | arm64 | Apple M1 (Virtual) | 3.12.10 | 1.96.1 | macos15-20260907.0337.1 |
| Ubuntu 24.04 · arm64 | Linux-7.0.0-1017-azure-aarch64-with-glibc2.39 | aarch64 | Neoverse-N2 | 3.12.10 | 1.96.1 | ubuntu24-arm64-20261018.154.1 |
| Ubuntu 24.04 · x86_64 | Linux-6.17.0-1022-azure-x86_64-with-glibc2.39 | x86_64 | AMD EPYC 7763 64-Core Processor | 3.12.10 | 1.96.1 | ubuntu24-20261012.334.1 |
| Ubuntu 24.04 · x86_64 · Pillow-SIMD paired cohort | Linux-6.17.0-1022-azure-x86_64-with-glibc2.39 | x86_64 | AMD EPYC 7763 64-Core Processor | 3.12.10 | 1.96.1 | ubuntu24-20261012.334.1 |
Comparison cohorts¶
| Runner | Cohort | Run | Measured at | Source SHA-256 |
|---|---|---|---|---|
| macOS 15 · arm64 | default | migration-benchmark-2d602fc00d24492d854862d10a5d5989 | 2026-10-21T11:26:29.269699Z | 10d57e5ad6af557d693c648420dc107c7a35b10f6b91ad65a7084a8955e66360 |
| macOS 15 · arm64 | parallel-cpu | migration-benchmark-f26cf18101b447b091c011e1133d2726 | 2026-10-21T11:34:28.497946Z | 11a70bf277b23f270217a7269b4b12c61b6aa1a19ca531df352cb3ffb2d9b3ac |
| Ubuntu 24.04 · arm64 | default | migration-benchmark-23fb0d6397ec4c77aae7c3bee551af0a | 2026-10-21T10:40:04.393859Z | 226c9e4d40d080e8a5ebc76d5252ac5650f7edb18675b1ffd83904ecd7748b6a |
| Ubuntu 24.04 · arm64 | parallel-cpu | migration-benchmark-e55a71e76b1e4f8e8b6ca2429855e664 | 2026-10-21T10:45:57.154575Z | cbef9444af5a38f4609597377d1aa6ea1b89d138c16b84bec6f19d936d01c4a2 |
| Ubuntu 24.04 · x86_64 | default | migration-benchmark-da602e6ca6e94ecfb5bcba62d14913e2 | 2026-10-21T10:46:35.220430Z | eedf096239320291c9ad6cbb42f8fa20ae3cc2542f88f50373be7faacda4e5ab |
| Ubuntu 24.04 · x86_64 | parallel-cpu | migration-benchmark-75fe6831874f4df6a7fb86d7f5eac0ae | 2026-10-21T10:53:26.181116Z | 10314424e0b883b36aa69d39e731385bf1ff516993d73c5e73e80c45ba1f1d3e |
| Ubuntu 24.04 · x86_64 · Pillow-SIMD paired cohort | pillow | migration-benchmark-15a619ce4f8547ff84fb69e2fa482c31 | Not measured | Not measured |
| Ubuntu 24.04 · x86_64 · Pillow-SIMD paired cohort | pillow-simd | migration-benchmark-231f0c51d0f845048dcc2e801d85d6d7 | Not measured | Not measured |
Download the public measurement data. This is a presentation snapshot, not the full source receipt. It preserves the original report hash and numerical observations; hostnames, local paths, and internal traces are omitted.
- Whole-workflow timings. successful_execution is an execution gate, not a pixel-parity claim.
- Requested and actual backends are separate. Host controls and missing terminal evidence do not prove native GPU performance.
- Parallel CPU is an opt-in Rayon build measured separately. Its parity preflight uses Pillow, while its speed ratios reuse ordinary Pillow timings from the default run; no second Pillow timing is collected.
- Each runner cohort is a separate GitHub-hosted measurement. Ratios are formed only between matching rows from that same cohort; runner cohorts are not compared directly.
- The reader summary counts faster, slower, and tied parity-verified workloads separately by runner and operation/pipeline scope; counts are not production-traffic weights or confidence estimates.
- The expandable per-workload ratio plot shows each parity-verified workload as a labeled observation; it is not a confidence interval or a universal speed guarantee.
- This x86-only cohort compares pillow-rs CPU/SIMD, ordinary Pillow 12.1.1, and Pillow-SIMD 12.1.1.post0 on identical workloads and measurement policies.
- Pillow-SIMD and Pillow were run in separate virtual environments because both install the PIL namespace. pillow-rs CPU/SIMD parity uses regular Pillow; Pillow-SIMD compatibility is checked separately and incompatible baseline cells remain visible with their speed ratios withheld.
- The matched cohort covers 37 full-size, parity-backed workflow IDs: 32 pipeline-op workloads, the established RGB GaussianBlur case, and getchannel in L, LA, RGB, and RGBA. Each measured source uses the same input and measurement policy per workflow; target profiles appear only when declared by that workload.
- This cohort uses the Pillow-SIMD default SSE4 build. It is not comparable with the Apple ARM measurements on the main benchmark page.
- Pillow-SIMD output differed from regular Pillow on 3 workload(s); those baseline cells are retained and no Pillow-SIMD speed ratio is accepted: pipeline-op.filter.material-l-noise-5x5-1024x768, pipeline-op.filter.material-la-nonmatching-5x5-1024x768, pipeline-op.grayscale.material-rgb-noise-1024x768.
- Pillow-SIMD rows are paired with ordinary Pillow and pillow-rs CPU/SIMD from one Linux x86 benchmark run; that narrow cohort is kept separate from the full Linux and macOS workloads.
Scope¶
The linked snapshot retains each recorded workload, context, timing, correctness result, and backend receipt. Its workload set is historical and does not represent the current generated public-operation inventory. Use it for the recorded run only; do not interpret its aggregate counts as current API coverage.
Correctness, native dispatch, timing regressions, process memory, and artifact size are separate claims.