H1: make bench, render budgets, and a third of the figure draw calls

make bench (tools/bench.sh, tools/bench_report.gd) measures the server's
p50/p99/max tick with 24 bots on three seeds over a whole match, and the
client's draw calls, primitives, objects, nodes and memory at six moments
of a match rendered in the capture container. It fails on anything over
the budgets now recorded in docs/ARCHITECTURE.md: p99 tick 12 ms, 2,000
draw calls and 2,000,000 primitives per frame (all passes), 1 GiB video
and 512 MiB static memory. Every capture now reports its frame's render
counts in a "render" field.

The first run put the barge, with 24 figures together, at 2,197 draw
calls. Each figure part had one surface per colour (about 43 per figure,
drawn again for shadows). FigureRig now merges a part's surfaces into one
per finish (coat, gilt, plain) and carries the Palette colours as vertex
colours, with shared materials. The barge is down to 1,207 draw calls,
with the same triangles and the same look.

Frame rates on a real GPU are not measured here; that stays with R6.

Closes #54

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-13 23:30:16 -05:00
co-authored by Claude Opus 5.5
parent 0e6e12cbc8
commit 9ce0865c9b
9 changed files with 340 additions and 16 deletions
+23 -1
View File
@@ -518,8 +518,30 @@ client is connecting, in the lobby or watching.
| Server tick | Under **12 ms** with 24 players, including the network's share (`make soak` checks the p99) |
| Client | **60 fps** on a GTX 1060-class GPU at medium settings |
| Terrain streaming | Under **8 ms** of main-thread work in any frame (measured headless; GPU upload unmeasured) |
| Client draw calls | At most **2,000** per frame, every pass counted (shadows included) |
| Client primitives | At most **2,000,000** triangles per frame, every pass counted |
| Client memory | At most **1 GiB** of video memory and **512 MiB** of the engine's static memory |
Software rendering in CI can't verify the client budget. It needs a real GPU playtest.
Software rendering in CI can't verify the frame-rate budget. It needs a real GPU playtest
(R6). What it can count is what a frame asks the GPU to do, so the draw-call, triangle
and memory budgets stand in for it until then. They are set for the GTX 1060 class with
room left for effects and the HUD.
`make bench` (`tools/bench.sh`, H1) measures both sides and fails when a number is over
budget:
- **Server:** the headless server with 24 bots, unthrottled, for 12,000 ticks (a whole
match and the start of the next) on three seeds, reporting p50, p99 and max tick time.
The max includes one long tick per match: unthrottled, the 15 s results screen passes
in a fraction of a second, so the main thread waits for the worker still building the
next island. At the real 30 Hz the worker finishes long before the lobby opens.
- **Client:** the match `make match` saved, rendered at the whole island, the barge (24
figures together, the worst case for draw calls), landed, a build and the final
players, and a crowded pine forest. Each capture reports its frame's draw calls,
primitives, objects, nodes and memory (`tools/capture.gd`'s `render` field).
Results go to `build/bench/report.txt` and `report.json`. The numbers from each pass are
in the DEVLOG.
## Soak testing