Skip to content

Latest commit

 

History

History
108 lines (82 loc) · 3.48 KB

File metadata and controls

108 lines (82 loc) · 3.48 KB

Benchmarks

This page reports 54 benchmark combinations across three scenes, three display profiles, and six rendering modes. Measurements are available for 45 combinations. The three 8K MVR HR entries and six 8K Optical/Viewpoint entries are listed as TBA.

Test system

  • CPU: AMD Ryzen 7 7800X3D
  • GPU: NVIDIA GeForce RTX 5080
  • Memory: 64 GB RAM

Settings

  • Native output: 2K 1536x2048, 4K 3840x2160, and 8K 7680x4320
  • Scenes: Lego, Garden, and Bicycle in SOG format
  • Multiview Raster: alpha_threshold=1/255, SH degree 3, and the profile's native view count
  • Ray Tracing: t_threshold=0
  • VSync: off
  • Process: one fresh application process per combination
  • Warmup: 6 frames for MVR and Standard; 600 frames for reuse modes
  • Measurement: 12 timed frames per run, 3 runs per combination, reported as the median FPS
  • Precision: two decimal places

Reuse maps

Profile Mode Source count Actual beta
2K Optical Reuse 3,221,933 1.0242249171
2K Viewpoint Reuse 3,753,540 1.1932182312
4K Optical Reuse 12,296,078 1.4824553916
4K Viewpoint Reuse 12,426,730 1.4982072242
8K Optical Reuse TBA TBA
8K Viewpoint Reuse TBA TBA

Results

The columns are ordered as MVR LR, MVR MR, MVR HR, ROR Standard, Optical Reuse, and Viewpoint Reuse.

2K — 1536×2048

Scene MVR LR MVR MR MVR HR ROR Standard Optical Reuse Viewpoint Reuse
Lego 64.30 55.81 36.92 32.22 86.02 75.47
Garden 13.98 12.13 8.30 11.09 27.54 24.48
Bicycle 14.33 12.41 8.31 9.12 21.57 19.34

4K — 3840×2160

Scene MVR LR MVR MR MVR HR ROR Standard Optical Reuse Viewpoint Reuse
Lego 46.59 30.89 11.98 6.54 16.27 15.88
Garden 9.91 8.41 5.09 5.57 10.53 10.45
Bicycle 10.80 8.88 5.17 4.97 8.61 8.53

8K — 7680×4320

Scene MVR LR MVR MR MVR HR ROR Standard Optical Reuse Viewpoint Reuse
Lego 30.10 7.41 TBA 1.94 TBA TBA
Garden 0.65 0.56 TBA 1.63 TBA TBA
Bicycle 0.56 0.55 TBA 1.45 TBA TBA

Reproduce the measurements

Download the scenes and generate the 2K and 4K benchmark maps:

$BuildRoot = Join-Path $env:TEMP 'DirectLStudio-build'
$MapsRoot = Join-Path $env:TEMP 'DirectLStudio-benchmark-maps'
$BenchmarkRoot = Join-Path $env:TEMP 'DirectLStudio-benchmark-v1.0.0'

python .\tools\download_assets.py --scene all --output .\assets\scenes
python .\tools\benchmark.py prepare-maps --output $MapsRoot

Run the available matrix entries:

python .\tools\benchmark.py run `
  --exe (Join-Path $BuildRoot 'bin\DirectLStudio.exe') `
  --assets .\assets\scenes `
  --maps $MapsRoot `
  --output $BenchmarkRoot `
  --runs 3 `
  --frames 12 `
  --warmup 6 `
  --reuse-warmup 600 `
  --forbid-process lfs-benchmark.exe

Run a single combination with --case scene:profile:mode:

python .\tools\benchmark.py run `
  --exe (Join-Path $BuildRoot 'bin\DirectLStudio.exe') `
  --assets .\assets\scenes `
  --maps $MapsRoot `
  --output $BenchmarkRoot `
  --case lego:4K:optical

Valid mode names are mvr_lr, mvr_mr, mvr_hr, ror, optical, and viewpoint.

Export the published table as JSON:

python .\tools\benchmark.py published `
  --output (Join-Path $BenchmarkRoot 'published-54.json')