● live2025CASE STUDY 04 / 07

Bottleneck Calculator

Visualise CPU/GPU bottlenecks and get upgrade suggestions for balanced PC builds.

ReactViteTailwind CSSFastAPIPython

1,434

CPUs scored

0–10000

Score scale

S–D

Tiers

85%

Cinebench imputed

bottleneck.tech-latest.com
Bottleneck Calculator screenshot

01 / The problem

Min-max normalized scores broke every time a new flagship GPU shipped — when the RTX 5090 joined the dataset, its outlier score compressed every other card's rank even though nothing about them had changed.

02 / The approach

Switched to percentile-based ranking, which is stable against new outliers, combined with geometric-mean weighting across benchmark sources so a component has to be consistently strong — not just spike in one test — to score well.

03 / The catch

Percentile scores are rank-preserving but not ratio-preserving. Feeding them directly into FPS Calculator's CPU/GPU ratio math produced absurd predictions — 350 FPS in Valorant for a GTX 1060 — so a separate ratio-preserving raw score had to be exposed just for that math.

Web-based tool that analyses CPU/GPU combinations, computes bottleneck percentages, and recommends balanced upgrades. React + Vite + Tailwind frontend backed by a FastAPI Python service serving calibrated JSON benchmark datasets. Shares the same benchmark API layer as FPS Calculator.

Highlights

1,434 CPUs scored; ~85% missing Cinebench data imputed via regression (R²≈0.90)

Percentile normalization survives new-hardware outliers without reshuffling every existing score

Same benchmark API powers FPS Calculator's predictions — one shared scoring engine across both sites

Upgrade suggestions tuned per resolution and workload (gaming / productivity / overall)

Gallery

Like what you see?

Let's build something together.