alint benchmarks: retired 3900X series
Hardware fingerprint: linux-x86_64-ryzen-3900x · AMD
Ryzen 9 3900X (12-core) · 62 GB · ext4 · rustc 1.95. Absolute
numbers are not comparable across machines; kbench is ~1.5× slower
per-core, which is why the current series was re-baselined rather
than spliced onto this one. See the
methodology document
for what does and doesn't transfer.
Full per-release trajectory (3900X)
1M-file headline cells across the most-stressed scenarios, every published release from v0.5.6 to v0.13.0. S3 is the realistic- monorepo anchor; S6 is the per-file content fan-out; S7 is the cross-file relational cliff that v0.9.8 closed; S9 is the nested- polyglot scenario.
| Version | Date | 1M S3 full | 1M S6 full | 1M S7 full | 1M S9 full |
|---|---|---|---|---|---|
| v0.13.0 | 2026-06-17 | 11.82 s | 11.57 s | 15.41 s | 7.48 s |
| v0.12.0 | 2026-06-07 | 12.00 s | 10.89 s | 15.55 s | 7.63 s |
| v0.11.0 | 2026-05-28 | 12.05 s | 11.99 s | 15.12 s | 7.70 s |
| v0.10.2 | 2026-05-21 | 12.28 s | 12.27 s | 15.50 s | 7.94 s |
| v0.10.1 | 2026-05-20 | 11.91 s | 12.02 s | 15.51 s | 7.96 s |
| v0.10.0 | 2026-05-20 | 11.82 s | 12.14 s | 15.55 s | 7.70 s |
| v0.9.23 | 2026-05-17 | 11.94 s | 11.88 s | 15.23 s | 7.55 s |
| v0.9.22 | 2026-05-14 | 11.71 s | 11.74 s | 15.40 s | 7.41 s |
| v0.9.21 | 2026-05-14 | 11.60 s | 11.75 s | 15.40 s | 7.64 s |
| v0.9.20 | 2026-05-10 | 12.38 s | 12.20 s | 15.30 s | 7.88 s |
| v0.9.19 | 2026-05-09 | 12.15 s | 11.79 s | 15.38 s | 7.60 s |
| v0.9.18 | 2026-05-08 | 11.70 s | 11.19 s | 15.56 s | 7.47 s |
| v0.9.17 | 2026-05-06 | 11.67 s | 11.38 s | 15.37 s | 7.37 s |
| v0.9.16 | 2026-05-06 | 11.58 s | 10.72 s | 15.35 s | 7.21 s |
| v0.9.14 | 2026-05-05 | 12.06 s | 11.19 s | 15.31 s | 7.33 s |
| v0.9.13 | 2026-05-04 | 11.46 s | 11.18 s | 15.45 s | 7.22 s |
| v0.9.12 | 2026-05-03 | 11.98 s | 11.33 s | 15.36 s | 7.46 s |
| v0.9.11 | 2026-05-03 | 11.84 s | 11.91 s | 17.29 s | 8.50 s |
| v0.9.10 | 2026-05-03 | 11.62 s | 11.22 s | 15.50 s | 7.21 s |
| v0.9.9 | 2026-05-03 | 13.23 s | 11.94 s | 17.32 s | 7.91 s |
| v0.9.8 | 2026-05-02 | 11.33 s | 10.89 s | 15.41 s | 7.32 s |
| v0.9.7 | 2026-05-02 | 11.89 s | 11.35 s | 614.4 s | 7.36 s |
| v0.9.6 | 2026-05-02 | 11.09 s | 11.40 s | 623.7 s | 7.12 s |
| v0.9.5 | 2026-05-01 | 12.59 s | 11.85 s | 652.4 s | n/a |
| v0.9.4 | 2026-04-30 | 731.9 s | n/a | n/a | n/a |
| v0.5.7 | 2026-04-26 | n/a | n/a | n/a | n/a |
| v0.5.6 | 2026-04-26 | 569.1 s | n/a | n/a | n/a |
The two cliffs this series captured
The dramatic version-over-version stories that shaped alint's engine, both bolded in the table above. Each shipped with an investigation write-up documenting the diagnostic data, the bisect, and the fix.
- S3 1M, v0.9.4 → v0.9.5: 731.9 s → 12.59 s. The lazy path-index + literal-path fast paths. Before the fix, the realistic-monorepo workload was quadratic in the path-prefix check; after, linear.
- S7 1M, v0.9.7 → v0.9.8: 614.4 s → 15.41 s. The
cross-file dispatch fast paths, round 2: the
pair/unique_by/for_each_dirrelational kinds made linear in the file count.
Full write-ups:
investigations/.
The raw per-release snapshots for this series live under
macro/results/linux-x86_64-ryzen-3900x/.
Looking for current numbers? They're measured on the dedicated
kbench host, per release.