mirror of
https://github.com/horizon-rust/horizon.git
synced 2026-11-07 11:06:11 -10:00
benchmarks: label the 2026-11-04 Mac mini runs offscreen, not comparable
Horizon ran offscreen (no presenting) while vanilla drew into a window, so the Mac rows are not a fair comparison; the website keeps the Mac mini's 2026-10-22 runs. The reports stay as a record, labelled. Co-Authored-By: Claude Opus 5.5 <[email protected]>
This commit is contained in:
1 parent
be11ce8db1
commit
a793d0b878
1 file changed
+8
-4
@@ -1,6 +1,10 @@
|
||||
# Horizon Nightly 2026-11-04 vs vanilla 26.3 on the Mac mini and the desktop (2026-11-04)
|
||||
|
||||
The runs behind playhorizon.dev/benchmarks as of 2026-11-04 (horizon-rust/horizon-site#58): `tools/perf_bench.py`
|
||||
> **The Mac mini reports (`*-macmini-rd*.json`) are offscreen, not comparable.** Horizon rendered offscreen (no presenting)
|
||||
> while vanilla drew into a window, so the Mac rows below are not a Horizon-vs-vanilla comparison, and the website does not
|
||||
> publish them: it keeps the Mac mini's 2026-10-22 runs. They are kept here only as a record of that day's offscreen run.
|
||||
|
||||
The desktop runs are the ones behind playhorizon.dev/benchmarks as of 2026-11-04 (horizon-rust/horizon-site#58): `tools/perf_bench.py`
|
||||
reports, render distance 12 and 32, three games each, every game from a fresh world (seed 1234), 1280x720, VSync off. Files:
|
||||
`{horizon,vanilla}-macmini-rd{12,32}.json` and `{horizon,vanilla}-desktop-rd{12,32}.json` next to this page.
|
||||
|
||||
@@ -13,7 +17,7 @@ reports, render distance 12 and 32, three games each, every game from a fresh wo
|
||||
|
||||
## Machines and conditions
|
||||
|
||||
- **Mac mini, Apple M4 Pro (14 cores, 20-core GPU), macOS.** The session was locked, so Horizon ran `--offscreen`: a locked
|
||||
- **Mac mini, Apple M4 Pro (14 cores, 20-core GPU), macOS: offscreen, not comparable.** The session was locked, so Horizon ran `--offscreen`: a locked
|
||||
session gives a window no frames, and windowed Horizon measured 0 FPS. Vanilla drew into its window. Offscreen skips
|
||||
presenting, which favours Horizon, so its averages aren't held at the 144 Hz display as they were on 2026-10-22. Each game
|
||||
started at a 1-minute load of 17-18 (the run waited for < 20); during the runs the load rose to 30-50.
|
||||
@@ -29,8 +33,8 @@ reports, render distance 12 and 32, three games each, every game from a fresh wo
|
||||
|
||||
| Machine | RD | Horizon | Vanilla | Horizon ÷ vanilla (avg, 1% low) |
|
||||
|---|---|---|---|---|
|
||||
| Mac mini | 12 | 774 / 294 (offscreen) | 132 / 27 | 5.9x, 10.9x |
|
||||
| Mac mini | 32 | 297 / 159 (offscreen) | 68 / 16 | 4.4x, 9.9x |
|
||||
| Mac mini (**offscreen, not comparable**) | 12 | 774 / 294 | 132 / 27 | — |
|
||||
| Mac mini (**offscreen, not comparable**) | 32 | 297 / 159 | 68 / 16 | — |
|
||||
| Desktop | 12 | 2,224 / 308 | 1,824 / 201 | 1.2x, 1.5x |
|
||||
| Desktop | 32 | 800 / 140 | 696 / 64 | 1.15x, 2.2x |
|
||||
|
||||
|
||||
Reference in new issue
Block a user