1# llimphi-gpu-bench23Standalone binary to validate the "direct wgpu GPU" section of
4`02_ruway/llimphi/SDD.md` on a machine with a real GPU.
56It does four things in order and prints everything to stdout in
7copy-paste-friendly markdown/table form:
891. **System header** — version, time, OS, detected GPU.
102. **wgpu adapter info** — backend (Vulkan/Metal/DX12/GL), device name, vendor,
11 relevant limits.
123. **vello vs direct GPU spike** — for N ∈ {25K, 50K, 100K, 200K, 500K, 1M}. It
13 measures ms/frame for each and the factor, and evaluates the SDD's criterion:
14 ≥5× at 500K → PASS; below → ABORT.
154. **Direct-GPU scaling alone** — for N ∈ {100K, 500K, 1M, 2M, 5M, 10M}. It
16 measures ms/frame, equivalent fps and Mprim/s, and evaluates the 60 fps @ 1M
17 goal.
185. **Visual verification PNGs** — it exports two files to the cwd:
19 `bench_vello_100k.png` and `bench_directo_100k.png`. The shape of the dot sky
20 must match between the two (deterministic LCG).
2122## Use2324```sh25cargo run --release -p llimphi-gpu-bench
26```2728---
2930Part of **llimphi** — see [llimphi](../README.md).
README.md
llimphi-gpu-bench
Standalone binary to validate the "direct wgpu GPU" section of
02_ruway/llimphi/SDD.md on a machine with a real GPU.
It does four things in order and prints everything to stdout in
copy-paste-friendly markdown/table form:
vello vs direct GPU spike — for N ∈ {25K, 50K, 100K, 200K, 500K, 1M}. It
measures ms/frame for each and the factor, and evaluates the SDD's criterion:
≥5× at 500K → PASS; below → ABORT.
Direct-GPU scaling alone — for N ∈ {100K, 500K, 1M, 2M, 5M, 10M}. It
measures ms/frame, equivalent fps and Mprim/s, and evaluates the 60 fps @ 1M
goal.
Visual verification PNGs — it exports two files to the cwd:
bench_vello_100k.png and bench_directo_100k.png. The shape of the dot sky
must match between the two (deterministic LCG).