Case study 04 / 07 · ● Live
Bottleneck Calculator
Visualise CPU/GPU bottlenecks and get upgrade suggestions for balanced PC builds.
A web tool that analyses CPU/GPU pairings, computes bottleneck percentages, and recommends balanced upgrades, with a React frontend and a FastAPI backend serving calibrated benchmark datasets.
- React
- Vite
- Tailwind CSS
- FastAPI
- Python
Min-max normalized scores broke whenever a new flagship GPU such as the RTX 5090 was added: its outlier score shifted the ranks of every other card.
Switched to percentile-based ranking, which absorbs new outliers without reshuffling existing scores, plus geometric-mean weighting across benchmark sources so a component has to perform consistently rather than spike in one test.
Percentile scores preserve rank but not ratios, which produced unrealistic FPS predictions (about 350 FPS in Valorant for a GTX 1060). A separate ratio-preserving raw score is exposed for FPS Calculator’s math.
Highlights
- 1,434 CPUs scored, with about 85% of missing Cinebench data imputed by regression (R² ≈ 0.90)
- Percentile normalization that stays stable when new hardware is added
- One shared scoring engine powers both this tool and FPS Calculator
- Upgrade suggestions tuned by resolution and workload (gaming, productivity, or overall)
Like what you see?