Dino SlothIndependent performance review · DS/PERF/01
Executive excerpt · fictional data
Project Firebreak
Performance position and evidence from a repeatable three-scenario review.
- Build
- 0.8.41 · Development
- Target
- 60 fps · console
- Coverage
- Three test routes
- Confidence
- Moderate
60 fps not yet demonstrated
Ordinary play is close. Combat and the first use of key UI break the frame budget.
The worst repeatable route is CPU-limited; a separate GPU spike appears when the storm effect fills the screen. The evidence does not support an engine rewrite. Remove the synchronous load, budget the effect and repeat the route on target hardware.
01 · Measured position
Median frame time by scenarioLower is better · target 16.67 ms
Training room13.8 ms
Market encounter20.9 ms
Storm approach18.7 ms
Vertical marker = 16.67 ms frame budget. Two cold runs per scenario.
34.8 msWorst repeatable p99
4.6 msFirst-use CPU cost
5.1 msStorm GPU cost
8.7 GiBPeak observed memory
Dino SlothProject Firebreak · Finding 01
Worked finding · CPU / loading
High priorityHigh confidence
First use of the equipment wheel blocks the game thread on thumbnail loads.
The first open during the market encounter produces a 29–35 ms frame. Later opens in the same run do not. Two cold captures align the hitch with synchronous package-load events; preloading the same thumbnails removes 4.2–4.6 ms and the visible spike.
Player effect
A visible hitch occurs during the first combat tutorial.
Recommended change
Use the existing asynchronous asset path and cache the resolved set.
Closure test
Cold boot; first-open p99 below 20 ms on target hardware.
02 · Recommended sequence
- Now
Remove the first-use load hitchSmall, visible and directly supported by the captures.
2–4 days
- Now
Put the storm effect on a GPU budgetProfile emitters separately; agree the visual answer with technical art.
2 TA days
- Next
Repeat the route on target consoleThe PC proxy identified suspects; it cannot provide console sign-off.
1–2 days
- Next
Narrow the cook dependencyThen retain timing history to confirm the working loop improved.
3–5 days
03 · What not to do yet
Do not begin a broad animation, LOD or architecture rewrite.
Nothing in the reviewed evidence justifies one. If a measured cost leads into a specialist system, we work with the client's discipline owner or agree the appropriate specialist scope.
Review boundary: this excerpt does not establish long-session memory stability, multiplayer behaviour or final console performance.