Free, open-source 3D sculpting for Linux, macOS and Windows: dense meshes, remeshing, painting, built-in AI (image/text to 3D), Send to Unreal, game-ready LODs, Gaussian splats and OpenUSD.
C++ 3d3d-sculptingcppdear-imguidigital-sculptingfbxgame-developmentgaussian-splattinggltfimage-to-3dmcpopenglopenusdsculptingtext-to-3dunreal-engine
REPORT.md for feat/bench-public
islandmedb00cc4749f43
diff --git a/REPORT.md b/REPORT.mdindex 164bc9aa..d7e57574 100644--- a/REPORT.md+++ b/REPORT.md@@ -1,8 +1,8 @@-# site/membership+# feat/bench-public-- website/support.html (title "Membership") replaces donate.html; donate.html is a redirect (meta refresh + JS) keeping the query string, so the Stripe success URL donate.html?thanks=1 still works (checked in headless Chromium).-- Tiers are presented as memberships with their benefits; there is a One-time Supporter option (pay what you want, $5 minimum); the tax-deductible line is replaced with "Memberships are subscriptions to SculptForge benefits; cancel anytime in the customer portal".-- All site footers point to support.html with the label "Membership". Also updated: the index, studios, the guide (welcome, What's New), README nav, the sitemap, the in-app Help > Support, welcome, reminder and launch card, and the tests.-- Tests: build OK; ctest -j3 had 2 UI tests fail under load (ui_ctrl_brushes, ui_dyntopo); both pass on rerun.-- No .github/FUNDING.yml exists (not created). No Spanish strings contained donation wording.-- Commit: b3a8bc8+- Standard scenes: `sculptforge --benchmark standard-15m | standard-33m | standard`, and Help > Benchmark (the dialog defaults to Standard 15M, with 33M, both, or the old sphere sweep). The mesh is a procedural head (`makeBenchHead`, core/Benchmark.cpp) built from a formula, so every platform gets the same one: 15.05M / 33.2M triangles with subdivision levels. Strokes (`stroke_set` "standard/1"): Clay 25/60/120 px, Move 80 px (smaller loop), Smooth 60 px, Clay 240 px, then an orbit.+- New JSON fields: `settings.scene`, `stroke_set`; per run `worst_stall_ms`, `gpu_buffers_mb`, `vram_used_mb` (only when the driver reports it: NVX/ATI meminfo); per phase `brush`, `worst_stall_ms`. Move has no lag (it places no dabs).+- Measurement fixes: the lag match now only moves forward along the path (the figure eight crosses itself, which made p95 lag jump to about 500 ms / 2000 px). Proof mode now waits for the first-launch hardware calibration instead of running at the same time (it does a fresh calibration because settings are isolated), and the Welcome window stays hidden.+- website/benchmarks: rebuilt with the 3 run steps, a scene and stroke filter, a results table (machine with full hardware, OS, CPU, GPU, RAM, build plus commit link, fps, frame p50/p95/p99, lag px, worst stall, peak RAM, GPU memory, video), and how to submit: submit.py (new `--machine`) or the new **Benchmark result** issue form (label `benchmark` created). Says plainly that only SculptForge numbers are listed. Linked from every site footer and the sitemap. Docs: docs/benchmark.md, guide performance.md (guide HTML rebuilt), CHANGELOG [Unreleased].+- Measurement: NOT done. The load stayed between 8 and 30 for the full 30-minute wait, so no run and nothing added to results.json (it is still empty). Provisional 15M smoke numbers (1.5 s strokes, load about 20, 1690x2136 tiled window, M1 Pro, Asahi): clay S/M/L p99 32/36/75 ms; Move p50 27, p99 46; Smooth p99 49; 240 px p99 218 ms, worst stall 223 ms (releasing the 240 px stroke); peak RAM 2.1 GB, GPU buffers 209 MB. 33M was not run.+- Tests: ctest -j3 (Unix Makefiles, Release, nice/ionice): 116/116 passed. New test: benchmark_standard_head. submit.py selftest extended.