Skip to content

main-route-selected-hour-0_5m-base.md

Latest commit

 

History

History
81 lines (55 loc) · 5.26 KB

main-route-selected-hour-0_5m-base.md

File metadata and controls

81 lines (55 loc) · 5.26 KB
Raw

Main Route Selected-Hour 0.5m Base Baseline

Date: 2026-05-15

Scope

This artifact measures the main route /, not /debug. It captures the selected-hour WebGPU path at gridResolution=0.5 without .bin, Python, parity, or debug comparison data in the timing baseline.

JSON source: data/performance-results/main-route-selected-hour-0_5m-base.json

Included Analyses

  • Ben-Gurion/20250815_grid_2m_fullday
  • Ness-Tziona/exploded/nes_tziona_unblock_2

Excluded BG Variants

This 0.5m stress pass intentionally excludes other Ben-Gurion variants and uses only the BG base case plus the Ness Tziona base/exploded model.

Collection Method

  • Route: /
  • Query: gridResolution=0.5&utciRender=auto&utciRenderDiagnostics=1
  • Color modes: normalized/full-day and discrete/per-hour
  • Scrub sample: app-visible hour slider scrub from hour 0 to hour 1

Proof Boundary

  • Ben-Gurion/20250815_grid_2m_fullday: utciSurfaceSource=compute-buffer-selected-hour, baseRenderTransport=compute-buffer-selected-hour, dataTextureBuildCount=0, selectedHourRuntimeContract.strongVisibleGpuPath=true, no python/bin/debug comparison fields, no forbidden comparison requests.
  • Ness-Tziona/exploded/nes_tziona_unblock_2: utciSurfaceSource=compute-buffer-selected-hour, baseRenderTransport=compute-buffer-selected-hour, dataTextureBuildCount=0, selectedHourRuntimeContract.strongVisibleGpuPath=true, no python/bin/debug comparison fields, no forbidden comparison requests.

Memory is scoped only to tracked app-owned UTCI/WebGPU buffers. The GPU VRAM total mirrors the main-route runtime helper: persistentExposureBytes + allHoursOutputBytes + selectedHourOutputBytes + renderOwnedSelectedHourBytes. It is not a browser, OS, or device VRAM measurement.

Timing Table

Project Mode Phase Points First ready ms First visible ms Payload ms Worker BVH ms Upload ms Exposure precompute ms Dispatch ms Render update ms Scene sync delay ms Scene sync total ms
Ben-Gurion normalized initial 1,662,657 7781.2 8414.5 124.7 98.5 190.9 6055.4 4.8 2176.1 1767.0 408.7
Ben-Gurion normalized scrub 1,662,657 103.5 986.9 124.7 98.5 190.9 6055.4 69.7 916.8 707.9 208.6
Ben-Gurion discrete initial 1,662,657 6799.2 7483.0 155.5 87.0 199.3 6040.2 8.0 1266.4 864.1 402.1
Ben-Gurion discrete scrub 1,662,657 46.0 998.9 155.5 87.0 199.3 6040.2 11.6 987.1 785.1 201.8
Ness-Tziona normalized initial 8,171,761 20665.8 23883.8 277.4 215.9 1084.8 16224.2 16.4 7478.5 5707.2 1771.0
Ness-Tziona normalized scrub 8,171,761 182.6 4889.3 277.4 215.9 1084.8 16224.2 16.9 4871.8 3665.4 1206.0
Ness-Tziona discrete initial 8,171,761 18400.8 21560.6 286.9 209.7 1074.7 16265.8 15.3 5103.4 3473.3 1629.7
Ness-Tziona discrete scrub 8,171,761 160.8 4707.8 286.9 209.7 1074.7 16265.8 11.1 4696.5 3609.8 1086.6

Render Sync Detail

Project Mode Phase Layout build ms Storage init wait ms Buffer copy ms Queue drain ms
Ben-Gurion normalized initial 99.1 139.2 0.0 170.0
Ben-Gurion normalized scrub 101.3 38.6 0.0 68.7
Ben-Gurion discrete initial 116.3 122.7 0.1 162.9
Ben-Gurion discrete scrub 85.0 53.7 0.0 63.0
Ness-Tziona normalized initial 480.8 574.1 0.0 714.5
Ness-Tziona normalized scrub 539.1 147.3 0.1 519.0
Ness-Tziona discrete initial 470.9 529.0 0.0 627.3
Ness-Tziona discrete scrub 523.1 149.1 0.0 414.2

Memory Table

Project Points GPU VRAM MiB Persistent exposure MiB Selected-hour current MiB Render-owned selected-hour MiB Scope
Ben-Gurion 1,662,657 228.33 63.43 6.34 158.56 utci-owned-webgpu-buffers
Ness-Tziona 8,171,761 1122.22 311.73 31.17 779.32 utci-owned-webgpu-buffers

Current Optimization Inference

The 0.5m data keeps the UTCI selected-hour dispatch small: roughly 5-70 ms for BG and 11-17 ms for Ness Tziona across initial and scrub samples. That means UTCI math is not the primary bottleneck.

Initial load is dominated by exposure precompute and render publication:

  • BG initial: about 6.0 s exposure plus 1.3-2.2 s render update.
  • Ness Tziona initial: about 16.2 s exposure plus 5.1-7.5 s render update.

Scrubbing avoids the cold-start exposure work, but Ness Tziona still spends about 4.7-4.9 s before the new hour is visible. The scrub bottleneck is therefore render-side publication/synchronization, especially scene sync start delay and render-owned selected-hour storage handling.

The next optimization pass should split into two tracks:

  1. Scrub responsiveness: reduce render-owned storage churn, scene sync start delay, and queue drain cost.
  2. 0.5m cold start: tile or cache exposure/BVH/upload work so the first visible selected hour does not require one monolithic full-grid exposure pass.