Fr0zkaandClaude Opus 5.5
c38e683292
Six null-RHI traces on 55e20f4, no address-space cap. FI: first admission 0.385 s, LOD0-2 settled
1.943 s (1.555 s from admission; L97 2.109 s), UWorld::Tick max 6.7 ms (L97 21 ms). SW: 0.415 /
1.706 s (1.292 s from admission), max 12.0 ms. EndPhysics gap <= 2 ms. Peak RSS ~2.32 GiB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1
1.6 KiB
Linux 102
Six complete null-RHI Unreal traces on 55e20f4, with the L97 low-priority tile settings and tick logging enabled. Unreal was launched as taskset -c 0-11 timeout 1740s; it had no address-space cap. No build was needed: the VFHost plugin source matches this checkout and its module binary postdates the source.
| Stratum | First admission | LOD0–2 settled | From admission | UWorld::Tick p99 / worst max |
AVoxelWorld::Tick p99 / worst max |
EndPhysics gap max |
|---|---|---|---|---|---|---|
| FI | 0.385 s (0.382–0.388) | 1.943 s (1.922–1.951) | 1.555 s (1.541–1.566) | 4.8–6.6 / 6.7 ms | 4.2–4.8 / 5.1 ms | 1 ms |
| SW | 0.415 s (0.414–0.417) | 1.706 s (1.706–1.713) | 1.292 s (1.289–1.298) | 8.5–10.0 / 12.0 ms | 7.9–9.4 / 11.3 ms | 2 ms |
Times are medians across three complete runs, with ranges in parentheses. Tick p99 columns show the three-run range; max is the worst per-run maximum.
Compared with L97, median settle from first admission was 2.109 s / 21 ms max UWorld::Tick for FI and 1.316 s / 11 ms for SW. L102 measured 1.555 s / 6.7 ms for FI and 1.292 s / 12.0 ms for SW. The largest EndPhysics gap fell from 16 ms / 7 ms to 1 ms / 2 ms.
Peak Unreal RSS (VmHWM) was 2,378.7 MiB (2.32 GiB): FI runs were 2,370.5–2,378.7 MiB and SW runs 2,275.3–2,295.6 MiB. There was no L97-era executable or RSS record available; a separate old host/plugin build would be needed, so the memory delta is unknown.
Failed L102 runs: none. All six exited 0 and settled LOD0–2. Logs and per-run summaries are in Saved/Linux/l102/unreal-arrival/. No installs, commits, pushes, or stashes.