Page:
Display benchmarking
Pages
Action system
Algebraic constraint-solving
Code flags
Command language
Constraints
Constructions
Continuous integration
Demo project details
Display benchmarking
Display
Elements and observables
Engine references
Entities
Ganja.js
Geometric algebra
Gram matrix parameterization
Heterogeneous collections
Home
Impossolid
Interface
Language benchmarks
Language selection
Licensing
Logging
Numerical optimization
Serialization
Symmetry of constraints
Test problems
User stories
No results
7
Display benchmarking
Glen Whitney edited this page 2025-12-22 19:55:30 +00:00
To make sure the display stays responsive, we've been tracking how changes affect its performance.
Procedure
Build with --release. Test on Chromium with VSCodium, gedit, and the AMDGPU_TOP GUI running, and all other programs closed. In case it matters, keep the display prototype fully visible on the right half of the screen, with the AMDGPU_TOP plots visible on the left half. Start the turntable and then monitor AMDGPU_TOP stats for a minute or so.
Results
| Commit | Description | Graphics pipe | Browser CPU | Browser GFX | Browser VRAM | GPU power | Comments |
|---|---|---|---|---|---|---|---|
4afc820 |
Sort fragments while shading | 50-60% | 15-30% | 50% | 9 W | ||
4afc variant |
SPHERE_MAX_INTERNAL = 12 |
80-90% | 15-20% | 80-90% | 27 W | ||
4afc variant |
SPHERE_MAX_INTERNAL = 4 |
40-50% | 15-20% | 35-40% | 8-9 W | ||
e80adf8 |
Correct hit detection | 50-60% | 15-30% | 50% | 9 W | Texture pipe and efficiency arbiter are more like 45-60% | |
6db9f5b |
Only draw the top few fragments | 55-65% | 15-30% | 50-60% | 9 W | ||
6db9 variant |
LAYER_MAX = 8 |
35-50% | 15-25% | 38-45% | 9 W | ||
a4236a3 |
Dim sphere interiors | 55-65% | 15-25% | 50-60% | 94 MiB | 10 W |