ORIX Convert — end-to-end benchmarks
Generated file — do not edit. Written by
Scripts/bench.shon 2026-11-21T17:19:55Z.
Every row is one orix-convert convert invocation of the release build, measured with /usr/bin/time -l: the whole pipeline
(detection, planning, queue, engine, validation of every output, safe writer), process start-up included. Inputs are generated in code
(orix-convert dev seed) and are not timed. Default settings of each path unless the case names one; default concurrency of the queue
(one job per performance core, at most 8; video and LibreOffice jobs limited as in the app).
- Machine: Apple M5 Max, 18 cores, 48 GB, macOS 27.0
- This Mac is shared and was under load during the run. The 1-minute load average at the start of each case is in the table; with 18 cores, a load above 18 means the cases competed for processors. A case that started with a load above 36 was run a second time and both runs are listed. These are observations on a loaded machine, not best cases; nothing was repeated to compute a variance.
- Wall = elapsed time. CPU % = (user + system time) / wall, as reported by
timefor the tool and the helper processes it waited for. Peak RSS = maximum resident set size reported bytime -lfor theorix-convertprocess (memory of helper processes such as ffmpeg or LibreOffice is not included in that figure).
| Case | Run | Input | Exit | Wall (s) | Throughput | CPU % | Peak RSS (MB) | Output | Load avg (1 min) |
|---|---|---|---|---|---|---|---|---|---|
| PNG → JPEG | 1 | 100 PNG, 4000 × 3000 (12 MP), 820.5 MB | 0 | 4.98 | 20.1 images/s | 428 | 1346 | 100 file(s), 201.6 MB | 22.70 |
| PNG → WebP | 1 | 100 PNG, 4000 × 3000 (12 MP), 820.5 MB | 0 | 11.47 | 8.7 images/s | 593 | 2045 | 100 file(s), 53.6 MB | 21.77 |
| PNG → SVG (image.vectorize) | 1 | 100 PNG, 4000 × 3000 (12 MP), 820.5 MB | 1 | 19.42 | 5.1 images/s | 623 | 2228 | 99 file(s), 172.2 MB | 21.79 |
| HEIC → JPEG | 1 | 100 HEIC, 4000 × 3000 (12 MP), 75.8 MB | 0 | 5.07 | 19.7 images/s | 320 | 1042 | 100 file(s), 194.5 MB | 27.67 |
| 4K H.264 → HEVC (hardware encoder) | 1 | 3840 × 2160, 30 fps, 10 s, 36.5 MB | 0 | 3.54 | 2.8 s of video/s | 292 | 696 | 1 file(s), 35.5 MB | 27.06 |
| WAV → MP3 | 1 | 5 min, 44.1 kHz stereo 16-bit, 50.5 MB | 0 | 1.63 | 184.0 s of audio/s | 123 | 31 | 1 file(s), 7.5 MB | 26.65 |
| WAV → AAC (M4A) | 1 | 5 min, 44.1 kHz stereo 16-bit, 50.5 MB | 0 | 2.41 | 124.5 s of audio/s | 113 | 31 | 1 file(s), 6.8 MB | 26.65 |
| WAV → FLAC | 1 | 5 min, 44.1 kHz stereo 16-bit, 50.5 MB | 0 | 0.52 | 576.9 s of audio/s | 165 | 31 | 1 file(s), 8.9 MB | 25.72 |
| PDF → PNG, 150 dpi | 1 | 100 text pages, US Letter, 0.1 MB | 0 | 4.13 | 24.2 pages/s | 96 | 55 | 100 file(s), 48.1 MB | 25.72 |
| PDF OCR (pdf.ocr) | 1 | 20 page images at 150 dpi without text layer, 5.6 MB | 0 | 21.24 | 0.9 pages/s | 44 | 395 | 1 file(s), 5.9 MB | 25.66 |
| CSV → Parquet | 1 | 953.7 MB, 9 columns | 0 | 2.00 | 476.9 MB of CSV/s | 473 | 1136 | 1 file(s), 129.4 MB | 21.57 |
| DOCX → PDF | 1 | 16 KB DOCX made by the documents engine from generated Markdown (250 paragraphs, 10 tables) | 0 | 3.03 | 13.2 pages/s | 89 | 231 | 1 file(s), 0.4 MB, 40 pages | 21.57 |
| Folder → ZIP (archive.create) | 1 | 476.8 MB in 130 files: half repetitive text (very compressible), half pseudo-random bytes (incompressible) | 0 | 2.78 | 171.5 MB/s | 95 | 31 | 1 file(s), 144.1 MB | 21.04 |
| ZIP → folder (archive.extract) | 1 | ZIP of the same tree, 144.7 MB | 0 | 0.54 | 883.0 MB/s | 85 | 34 | 130 file(s), 476.8 MB | 20.96 |
14 measurement(s), 1 with a non-zero exit status.
Cases that did not end with status 0
The time of such a row covers the whole batch, including the files that failed; what the tool printed:
- PNG → SVG (image.vectorize) (run 1, exit 1):
Summary: 100 job(s) — 99 converted, 1 failed, 0 skipped — 847.4 MB in, 180.4 MB out — 19.3 sFAILED image-0009.png: The converted file failed verification. A contour leaves the viewBox.